Open AccessArticle

A Machine Learning Approach to Predict Site Selection from the Perspective of Vitality Improvement

Bin Zhao

¹,

Hao Zheng

^2,*

and

Xuesong Cheng

Shanghai Academy of Fine Arts, Shanghai University, Shanghai 200444, China

Architectural Intelligence Group, Department of Architecture and Civil Engineering, City University of Hong Kong, Hong Kong SAR, China

Author to whom correspondence should be addressed.

Land 2024, 13(12), 2113; https://doi.org/10.3390/land13122113

Submission received: 4 November 2024 / Revised: 3 December 2024 / Accepted: 4 December 2024 / Published: 6 December 2024

(This article belongs to the Section Land Innovations – Data and Machine Learning)

Download

Browse Figures

Figure 1
Study area. (a) Shanghai’s location in China; (b) Distribution of Shanghai’s 16 districts. "> Figure 2
Research framework. "> Figure 3
Distribution of POI data of 13 types of second category service facilities in Shanghai. "> Figure 4
Total of 344 CMPBs that meet the criteria. (a,b) They are CMPBs-POI, CMPBs nuclear density analysis. "> Figure 5
Three levels of buffer zones of CMPBs. (a–c) 500 M buffer, 1000 M buffer, 2000 M buffer. "> Figure 6
Matrix heat map of 13 types of Second Category service facilities and VVVs in the three levels of buffer zones of CMPBs. The significance level in the figure is represented as * p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001, and ns p > 0.05. "> Figure 7
Specific conditions of 13 types of Second Category service facilities and VVVs in the three levels of buffer zones of CMPBs. (a–m) Food services, Residential Services, Company Services, Shopping Services, Traffic Services, Financial Services, Hotel Services, Tourism Services, Life Services, Leisure Services, Education Services, Hospital Services, Government Services. "> Figure 8
ANN model for research settings. "> Figure 9
Training epoch and loss. "> Figure 10
Comparison of ANN model training with other models. "> Figure 11
VVVs prediction results of four districts in Shanghai, (a–d): Fengxian, Jiading, Qingpu, Songjiang. "> Figure 12
Comparison of scene grid scales. (a) The grid scale of the Shanghai Industrial Museum; (b) The grid scale of the Shanghai Museum North Building; (0–5) are SHP Plot File, Grid Overlay Analysis, Grid Scale: 5 m × 5 m, 10 m × 10 m, 20 m × 20 m, 50 m × 50 m. "> Figure 13
Shanghai Industrial Museum site situation and VVVs prediction results. (a–c) are site planning situation, research and design scope map, and design scope vitality value prediction. (a,b) the pictures provided by Shanghai Urban Planning and Design Institute were redrawn. "> Figure 14
Shanghai Museum North Building site situation and VVVs prediction results. (a–d) are site planning situation, research and design scope map, specific construction land schematic, design scope vitality value prediction. (a–c) are redrawn from the pictures provided by the Shanghai Museum. "> Figure 15
The technical route of combining a genetic algorithm and an ANN model. "> Figure 16
Grouping of genetic algorithm experiments (90 groups). "> Figure 17
The proportion of new VVVs greater than the original value. (a) “Second Category Service Facility Single Replacement” experiment. (b) “First Category Service Facility Classification Change” experiment. (c) “Replace All Service Facilities” experiment. "> Figure 18
MAX, AVG, MIN statistics of the proportion of new VVVs greater than the original value. "> Figure 19
Normalization analysis of all experimental groups except Scheme B. "> Figure 20
Box plot analysis of all experimental groups. "> Figure 21
The current situation of Songjiang Museum. (a) Satellite images of the surrounding environment. (b) Classification of land use. "> Figure 22
Numerical analysis of VVVs improvement in museums in Songjiang. (a) Comparing all experimental groups except Scheme B. (b) Comparison of “First Category Service Facility Classification Change” experiment. (c) Comparing the “Replace All Service Facilities” experiment. "> Figure 23
The total number of facilities allocated by Scheme B for different types of service facilities at different distances. "> Figure 24
Statistics on the allocation ratio of new service facilities of different types with different quantities and distances in Scheme B. ">

Versions Notes

Abstract

The selection of construction sites for Cultural and Museum Public Buildings (CMPBs) has a profound impact on their future operations and development. To enhance site selection and planning efficiency, we developed a predictive model integrating Artificial Neural Networks (ANNs) and Genetic Algorithms (GAs). Taking Shanghai as our case study, we utilized over 1.5 million points of interest data from Amap Visiting Vitality Values (VVVs) from Dianping and Shanghai’s administrative area map. We analyzed and compiled data for 344 sites, each containing 39 infrastructure data sets and one visit vitality data set for the ANN model input. The model was then tested with untrained data to predict VVVs based on the 39 input data sets. We conducted a multi-precision analysis to simulate various scenarios, assessing the model’s applicability at different scales. Combining GA with our approach, we predicted vitality improvements. This method and model can significantly contribute to the early planning, design, development, and operational management of CMPBs in the future.

Keywords:

cultural and museum public buildings; building site selection; development vitality; artificial neural network; genetic algorithm

1. Introduction

1.1. Research Background

Since the early 21st century, Shanghai’s urban construction boom has significantly increased the number of Cultural and Museum Public Buildings (CMPBs). However, this growth has led to operational challenges, with some CMPBs experiencing high visitor traffic and others relatively low attendance. These discrepancies arise from both subjective and objective factors. Subjective factors include the quality of management and the thematic focus of the CMPBs. A key objective factor is the construction site selection. Previous studies have shown that optimal site selection enhances the comprehensive functions of buildings and is crucial for their sustainable development [1,2]. The availability of public service facilities at different sites plays a crucial role in meeting residents’ daily needs, and significantly promote economic development and urban growth [3,4].

Our research suggests that for new and expansion projects, particularly those located far from the city center, it is essential to configure necessary public service facilities preemptively. This ensures that the project’s benefits are realized promptly after completion. CMPBs, which combine multiple services, require proper planning, construction, and configuration of related public service facilities to greatly enhance urban development. To minimize losses for both government and corporate developers, it is crucial to analyze site selection for CMPBs and investigate strategies to improve the vitality of existing CMPBs through systematic operational assessments.

1.1.1. Study Area

This study was conducted in Shanghai, China, a city known for its strong economy, abundant resources, and extensive social public service facilities (Figure 1). According to the Culture and Communication Dimension of the Global City Power Index1, Shanghai’s international ranking declined from 2015 to 2022. In response, Shanghai is striving to enhance its cultural soft power by promoting the construction of comprehensive cultural and sports facilities. Cultural and museum-type public buildings are essential for achieving this goal and play a significant role in promoting the high-quality development of Shanghai’s public cultural services.

1.1.2. Cultural and Museum Public Buildings (CMPBs)

CMPBs encompass not only traditional museums but also art galleries, exhibition halls, memorial halls, and science and technology museums, all of which have cultural collection and display functions. To facilitate this research, we will use the museum definition provided by ICOM and China’s JGJ66-2015 [5] Museum Building Design Code2. Based on these two key documents, CMPBs will be classified into four categories: art, history, science and technology, and comprehensive, considering both the nature of the collection and the content of the display (Table 1).

Research on CMPBs construction location prediction and vitality improvement using machine learning is relatively rare. Some studies use space syntax to analyze the spatial accessibility and location selection of CMPBs [6]. However, there are many studies on other building construction location predictions and building environment vitality optimization. From the perspective of typology and empirical analysis, other research areas provide important references for this study [7].

1.2. Problem Statement

Most current research on CMPBs construction site selection and vitality optimization focuses on large-scale, coarse-grained analysis [8,9]. This approach often results in inefficient site selection for CMPBs, leading to significant urban resource waste, particularly in terms of land, and hindering overall urban development. The results tend to be perceptual [10], lacking scientific and rigorous quantitative analysis. Improved CMPBs site selection can significantly promote urban development. Therefore, it is necessary to develop a comprehensive prediction model with multiple attributes. Such a model should meet the requirements for controllable accuracy, allow flexible switching between large and small urban scales, and support multiple prediction types. Coordinated use of these models can address the lack of coordination between current prediction methods and the results needed for macro-planning and refined construction of CMPBs.

1.3. Technical Discussions

1.3.1. Artificial Neural Network (ANN)

This research primarily employs deep learning-based analysis methods centered on Artificial Neural Networks (ANNs). ANN models excel in processing and analyzing vector data [11,12], and perform well even with limited or complex data sets by learning from large amounts of data [13]. Consequently, ANNs are widely utilized in architectural design and urban planning research [14,15,16]. ANNs demonstrate versatility and flexibility in urban spatial analysis, alongside excellent capabilities in fine-grained data processing and accurate data optimization, aligning well with the objectives of this study.

1.3.2. Genetic Algorithm (GA)

The Genetic Algorithm (GA) model is highly effective in solving complex optimization problems [17]. By integrating the optimization capabilities of the GA model with the evaluation strengths of the ANN model, more valuable research outcomes can be achieved in fields such as energy management, climate prediction, and engineering design [18,19,20]. In spatial analysis and architectural design, the combination of GA and ANN is particularly effective [21]. For instance, in urban planning, ANN models are used to learn and predict urban data, while GA optimizes urban planning parameters to achieve the best urban layout.

1.3.3. Model Comparison

Other deep learning models, such as Recurrent Neural Networks (RNNs), Convolutional Neural Networks (CNNs), and Generative Adversarial Networks (GANs), are not suitable for performing the tasks involved in this study [22,23]. Compared with ANNs, other networks, such as RNNs, CNNs, and GANs are more commonly used in tasks such as natural language processing (NLP), image classification, and real scene processing [24,25]. In addition, other traditional technological research methods are applied, such as using data advancement analysis models [26] and interactive multi-layer maps based on GIS systems [27], but with the rise of machine learning, new technological methods have higher efficiency and accuracy, which are worth exploring [28]. Therefore, from the perspective of enhancing development vitality, it is necessary to develop a prediction and optimization tool based on the relationship between urban service facilities and CMPBs construction site prediction. In this context, an ANN is the ideal choice for this study.

1.4. Objectives

To support sustainable development and efficient urban planning of CMPBs, our research introduces a machine learning-based prediction and analysis method. This study aims to achieve the following five objectives: Firstly, to conduct a correlation analysis of CMPBs in Shanghai using extensive POI data and representative VVVs data, enhancing the scientific accuracy of site selection decisions for future projects through data-driven insights. Secondly, to train and validate an ANN model through multi-scenario simulations and testing with untrained data, ensuring its reliability across various scales and scenarios. Thirdly, to propose an innovative model combining an ANN and a GA to predict the siting and vitality improvement of CMPBs, offering both theoretical value and practical utility for professionals. Fourthly, to develop an infrastructure optimization scheme for the Songjiang District Museum in Shanghai using this new method, demonstrating its effectiveness in enhancing the vitality and operational efficiency of CMPBs, and providing specific improvement suggestions based on actual conditions. Finally, to provide research results that inform government and institutional planning and policy for CMPBs, promoting rational and efficient resource allocation and offering researchers a novel approach.

2. Methodology

2.1. Research Framework

The ANN modeling framework for this study is as follows (see Figure 2): Firstly, we collected and processed Shanghai’s 2023 Points of Interest (POI) data, the city’s vector map, and the Visiting Vitality Data of CMPBs. We then organized the necessary POI data types and the number of CMPBs involved to establish the original database for building and training the ANN model. After training the model, we conducted an accuracy analysis to select a high-precision model as the final one. In the multi-scenario simulation, urban designers and related professionals can use this trained model to quickly predict the vitality of CMPBs construction sites in designated areas. Furthermore, the study integrates the trained ANN model with a Genetic Algorithm (GA) to devise optimal plans for enhancing infrastructure around the buildings, thereby boosting CMPBs’ vitality. Ultimately, this method not only improves the likelihood of successful CMPBs operations but also addresses urban issues related to land waste and misutilization.

2.2. Data Collection and Processing

2.2.1. POI, Shapefile Map, and VVVs Collection

Points Of Interest (POI) data were collected from the AutoNavi Open Platform3 by calling the AutoNavi Map API in November 2023. After preprocessing, irrelevant POI data, such as place name and address information, were removed. Ultimately, we used 986,474 POI data points. AutoNavi Map classifies POI data into 23 main categories and 265 secondary categories, including catering services, scenic spots, public facilities, shopping services, and scientific education and cultural services4. Our study is based on this classification system and references related studies [29,30]. We adjusted the classification to meet the research needs, focusing on three categories of first-level service facilities and 13 categories of second-level service facilities (Table 2 and Figure 3).

The Shapefile map data of Shanghai were collected through the Baidu Map API5 in December 2023, which included three levels of administrative area boundary maps (city, district, and township). The Shapefile map data of China come from the Resource and Environmental Science and Data Center of the Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, with the file collection time being 20236.

The Visiting Vitality Data for CMPBs were sourced from Dianping7, a popular social media platform for lifestyle information in China, known for its large user base and consumer reviews [31]. We collected CMPBs check-in data and the number of reviewers from December 2023 to January 2024, defining this data set as the Visiting Vitality Values (VVVs) of CMPBs.

2.2.2. Data Collection and Processing of the Three Levels of CMPBs Buffers

The data collection and processing process involved establishing buffer zones, connecting spatial data, and linking museum vitality data. Initially, we used the ArcGIS’s buffer tool to create three levels of CMPBs impact buffer zones with CMPBs as the center and radii of 500 meters (M), 1000 M, and 2000 M [32]. Next, using the ArcGIS’s spatial connection tool, we processed the 13 types of secondary service facility POI data within the three buffer levels. This resulted in 39 sets of data per object, with each type of secondary service facility having three rows of data corresponding to the 500 M, 1000 M, and 2000 M buffer zones. Additionally, we excluded CMPBs that could not provide valid VVVs, ultimately generating the input data for machine learning with a total of 344 valid CMPBs (Figure 4 and Figure 5).

2.2.3. Correlation Analysis of Data

To more accurately examine the differences in the buffer distances of CMPBs at the three levels, we analyzed the POI data of CMPBs and the POI data of 13 types of secondary facilities within the three buffer zones. We used Python to perform a correlation analysis on this data set, employing the Spearman correlation coefficient to assess the relationships between the buffer distances and the VVVs. The symbol

ρ

represents the correlation coefficient,

d_{i}

represents the difference in rank order of i,

n

represents the sample size (Equation (1)).

ρ = 1 - \frac{6 \sum d_{i}^{2}}{n (n^{2} - 1)}

(1)

The p-value, calculated using the Spearman correlation coefficient and sample size (N), plays a critical role in statistical analysis by assessing the significance of the observed correlation between variables. It represents the probability of obtaining a correlation at least as strong as the one observed in the sample, assuming there is no true correlation (null hypothesis). A smaller p-value suggests that the observed correlation is unlikely to have occurred by chance alone under the null hypothesis. Therefore, a smaller p-value provides stronger justification for rejecting the null hypothesis and concluding that there is a significant correlation between the variables being studied. The calculation of the p-value requires converting

ρ

into a

Z

value (Equation (2)).

Z = \frac{1}{2} \ln (\frac{1 + ρ}{1 - ρ})

(2)

The analysis results (Figure 6) show that, among the 39 data sets of 13 types of secondary facilities and VVVs at three buffer zone levels in Shanghai, the p-values for 38 data sets are less than 0.05. Only one data set has a p-value greater than 0.05, indicating that the correlation analysis for 97.4% of the data sets is statistically significant. The average values of the three-level buffer

ρ

are all positive values, being 0.20, 0.26, and 0.33, respectively. The numerical value of

ρ

reflects the strength of the relationship between the two variables. According to the numerical results, there is a moderate correlation between the 13 types of secondary facilities and VVVs, and as the distance increases, the correlation between the two gradually increases (Figure 7).

2.3. The ANN Architecture

TensorFlow, an open-source machine learning framework, is utilized extensively for constructing and training neural network models tailored for diverse data classification tasks [33,34]. In this study, TensorFlow is employed to develop an Artificial Neural Network (ANN) model comprising three main layers: an input layer with 39 neurons representing the service facility data for each Cultural and Museum Public Building (CMPB), four hidden layers with 80 neurons each (Figure 8), and an output layer consisting of one neuron representing the visitor vitality data for each CMPB. Generally, increasing the number of hidden layers and neurons within these layers has the potential to enhance prediction accuracy. While increasing the number of hidden layers and neurons can improve prediction accuracy, it also prolongs the training time and raises the risk of overfitting [35]. To address these challenges, we disabled TensorFlow’s Eager Execution mode in order to use the graph-based traditional execution mode, which may be easier to control and optimize in some complex neural network-training scenarios. For optimizer configuration, an AdamOptimizer optimizer was created with a learning rate of 0.001, and it was associated with the loss function cost using the minimize method to minimize the loss function during training.

Considering the training efficiency, training accuracy, and management risk, we selected the above specific parameters and then used the Sigmoid function to activate the results [34]. This process forms the fundamental operations that enable the network to learn and make predictions based on the input data. The calculation parameters are represented by

\hat{y}

, the neuron parameters are represented by

x

w e i g h t s

represents the weight of the layer, and

b i a s

represents the bias of the layer. The calculation process of the hidden layer output of each layer is (Equation (3)):

\hat{y} = S i g m o i d (w e i g h t s * x + b i a s) = \frac{1}{1 + e^{- (w e i g h t s * x + b i a s)}}

(3)

At the same time, to better verify the results of the output layer during the training of the ANN model, feedback on the deviation between the predicted output

\hat{y}

and the actual data output

y

is provided. The mean square error (MSE) function is selected in the setting of the loss function to evaluate the performance of the neural network, and back propagation is allowed to update the neural network parameters

w e i g h t s

and

b i a s

(Equation (4)). Thus, this study conducted an initial assessment of the ANN using the architecture described above, utilizing 300 training data sets to evaluate its training efficacy. The experiment comprised 42,000 training cycles, during which the model achieved its lowest error value of 0.0081 at the 41,400th cycle. Generally, a smaller value of the loss function indicates superior performance of the trained model (Figure 9).

L o s s (y, \hat{y}) = \frac{1}{n} \sum_{i = 1}^{n} {(y_{i} - \hat{y_{i}})}^{2}

(4)

2.4. The ANN Training

We reviewed pertinent literature on Artificial Neural Network (ANN) training and subsequently proposed our methodology [36]. The training was conducted on an AMD Ryzen 7 3700X 8-Core Processor CPU (3.60 GHz), with experiments exploring various training cycles (10,000, 20,000, 30,000, etc.) and optimizers (Adadelta, Adagrad, Adam, etc.). For optimal performance and efficiency, we determined that 42,000 training cycles were required, saving model parameters every 1000 cycles. We recorded progress updates every 100 cycles to monitor training progress. The specific training program mainly includes the following four parts:

Establishing the operating environment importing data. (1) To set up the environment in Jupyter, we begin by importing the necessary libraries. The csv module is used to process CSV (comma-separated values) files, which are commonly employed for structured data storage. The Numpy library is essential for managing data structures such as arrays and performing a variety of mathematical operations, making it crucial for data analysis and manipulation. Additionally, the math module provides basic mathematical functions, including logarithmic operations, which are integral to the calculations required for this project. These libraries collectively enable efficient data processing and computation within the Jupyter environment.; (2) To begin, we read the initial data by traversing the files in the directory. Once the files are loaded, we process the data by iterating through each entry. For each sub-list in the data, we apply a natural logarithmic transformation to the first element using the math.log function. This transformation is essential for normalizing the data and making it suitable for further analysis and modeling; (3) Next, two lists, maxi and mini, are initialized with lengths corresponding to the length of each sub-list in the data. Then, using two nested loops, the maximum and minimum values for each sub-list are identified. Afterward, the data are normalized by mapping them to a range of 0 to 1, ensuring consistency and scalability across the data set. Finally, the processed data are saved for use in subsequent analysis and modeling.
Design ANN architecture and set execution mode. (1) The model initialization function defines the input and output placeholders for the artificial neural network (ANN). These placeholders represent the features and labels that the model will use during training and prediction. Additionally, the function initializes the model’s weights and biases, which are essential parameters that the network will learn during training. This step is part of the overall structure of the ANN, as previously outlined in section “2.3”, and sets the foundation for the model’s learning process. (2) The multi-layer perceptron (MLP) structure begins with the input layer, represented by the input X, which consists of the feature data. The data are then passed through a series of hidden layers, where each layer performs a weighted sum of the inputs followed by an activation function. In this case, the Sigmoid activation function is used, which squashes the output of each layer into a range between 0 and 1, adding nonlinearity to the model. After passing through all hidden layers, the final output layer produces the model’s prediction. This output is the result of the MLP’s learned parameters (weights and biases), and is used for prediction or further analysis. (3) Define loss function and optimizer and set parameters to save the training model. (4) The model evaluation function assesses the trained model’s performance using a specified test data file. In the training environment, the saved model parameters are restored, and the predicted output for the given test data is generated and used for evaluation.
Design training programs and execute training. (1) A set of parameters related to the model training process are defined, including training_epochs, save_step, batch_step, display_step, and others. (2) The model is initialized, and the random seed is set by importing the random module from the Python standard library to generate random numbers. The seed is fixed at 3407 to ensure consistent random number generation each time the code is executed, thus ensuring repeatability. (3) A random index set is generated, containing all integers from 0 to the length of the data list minus one. From this set, 35 elements are randomly selected and stored in a list. This operation effectively samples about 10% of the data set. The selected elements are then sorted in ascending order, resulting in the output of the sampled index list. (4) The data set is divided into training and test sets, and the number of training samples is recorded. Four empty lists—”xdata” and “ydata” for storing training features and labels, and “X_test” and “y_test” for storing test data—are initialized. After processing the data, the final training sets “xdata” and “ydata”, the test sets “X_test” and “y_test”, and the number of training samples are printed. Next, the model is trained, and accuracy is calculated using the evaluation function. The model’s performance is assessed on the test data sets “X_test” and “y_test”, with metrics like accuracy being calculated and printed.
Test the performance of the model and make predictions. (1) A TensorFlow session is created to ensure that operations are carried out within the same computational graph and context. The model parameters are restored using the previously defined model saver object, which loads the trained model from the specified file. The model is then restored to the exact state it was at the specific checkpoint recorded in the save file. This allows subsequent prediction operations to be performed based on the model’s trained parameters, ensuring consistency and accuracy in the prediction process. (2) After restoring the model, the prediction operation is applied to the test data set, “X_test”. The model processes the test data, generating predicted output values. These predictions are then subjected to an exponential transformation to convert them back to their original scale or format, ensuring that the results align with the desired output form for further analysis or evaluation. (3) The sum of prediction errors is calculated. Through a loop, each element of the prediction output yhat is traversed, the absolute value of the error between the current prediction value yhat[i] and the corresponding true test value y_test[i] is calculated, and then a variable “error_sum” is initialized to 0 to accumulate the absolute value of the error between the predicted value and the true value. (4) The average error and accuracy are calculated. The previously accumulated error sum “error_sum” is divided by the number of samples of the predicted output “len(yhat)” to obtain the average prediction error value of each sample, which is stored in the “error_ave” variable. Then, according to the evaluation function, the indicators related to the accuracy are obtained to evaluate the prediction performance of the model on the test data set. (5) The prediction calculations and conversions are performed similarly to the previous steps, with the main difference being the use of a different data set for the calculation. In this case, the model’s prediction output operation is applied to the new test data set, “data_test”. The model processes this input to generate predicted outputs, which are then subject to any necessary post-processing, such as data transformations or conversions, to align with the expected output format. This ensures that the predictions made by the model can be accurately compared and evaluated against the true values of the new test data set.

During training, the ANN model’s performance evolves significantly across cycles. To determine the optimal stopping point, we systematically evaluated model accuracy at each epoch, selecting the iteration with the highest performance. Across multiple experiments, we documented three sets of training logs, each reflecting unique trends in accuracy improvement.

The first set, “Training 1”, initially shows inadequate accuracy for accurately predicting the construction site selection and VVVs optimization of CMPBs. However, as the training progresses, the accuracy steadily improves. By the 28th cycle, the model achieves stable high accuracy, reaching 92.9% by the 29th cycle. After the 39th cycle, slight fluctuations in accuracy prompted us to conclude the training.

In contrast, “Training 2” exhibits the most favorable results among the experiments, achieving 94.1% accuracy by the 29th cycle. Conversely, “Training 3” displays relatively poorer initial performance but maintains stability with minimal fluctuations over time. The prediction of construction site selection and VVVs optimization in CMPBs represents a complex system. In our results section, we conduct both qualitative and quantitative accuracy analyses to determine the most suitable model for scenario simulations and future applications.

Additionally, to enhance the robustness of our findings, we trained five additional deep learning models using the same data set. These models share a consistent operating logic and ANN structure, with a custom loss function designed to evaluate prediction accuracy. The custom loss function takes two parameters—ground truth and predictions—and computes the absolute error between them. This error is multiplied by 100, and the median of the resulting values is calculated (np. median). Finally, the result is rounded to one decimal place to provide a clear measure of accuracy. This function effectively quantifies the deviation between predicted and actual values, offering a standardized metric for evaluation. The median accuracy results obtained are as follows: DecisionTree Regressor 80.3%, Bagging Regressor 83.5%, RandomForest Regressor 83.7%, ExtraTrees Regressor 83.2%, and Linear Regressor 87.8%. Despite these competitive accuracies, the ANN model, particularly “Training 2”, surpasses these performances with 88.3% accuracy by the 18th cycle and peaks at 92.9% by the 36th cycle. This underscores the robust performance advantage of the ANN model in addressing the specific challenges of this research topic (Figure 10).

3. Results

3.1. AI-Assisted CMPBs Planning and Design

3.1.1. Large-Scale Regional Site Selection Prediction

We used the previous data set along with the ArcGIS fishing net tool to select four administrative districts in Shanghai. Based on the distance division criteria of the buffer zone for this study, we set the grid size for each administrative district to 1000 M by 1000 M, which corresponds to the pixel width and height in the fishing net tool. Additionally, we selected the “Create Annotation Points” option and chose “POLYGON” as the geometry type. After running the tool, we cropped the newly generated mesh to define the final study area grid. Next, we calculated the center points of each grid. To determine the latitude and longitude of these center points, we added fields, set the field type to floating point, used the “Calculate Geometry” function, and selected the X- and Y-coordinates of the center points based on latitude and longitude. We displayed the coordinates in the decimal system. After completing these steps, we obtained the central points with their corresponding latitude and longitude. Finally, we created three buffer zones (500 M, 1000 M, and 2000 M) based on the coordinate points of each grid center and calculated the number of POI data for 13 types of second-category service facilities within each buffer zone.

These data were imported into a trained ANN model for VVVs prediction. Using ArcGIS for visualization, we performed a normalized index processing of the predicted VVVs and applied the natural break point classification method to divide the results into five categories. Our study revealed that existing Cultural and Museum Public Buildings (CMPBs) are primarily located in high-VVVs areas predicted by the ANN model, confirming the model’s predictive accuracy. Additionally, we identified several high-VVVs areas without CMPBs, highlighting potential sites for future construction (Figure 11).

Specifically, Fengxian District is predicted to have the lowest VVVs for CMPBs among the four administrative districts. In contrast, Jiading District and Songjiang District are expected to have relatively high VVVs for CMPBs. In Jiading District, the high-VVVs points are evenly distributed, with existing CMPBs largely falling within these high-VVVs areas. In Songjiang District, the high-VVVs points are concentrated in the northeast and middle regions, where the existing CMPBs are also primarily located.

The situation in Fengxian District and Qingpu District differs significantly. In Fengxian District, high-VVVs points are rare and mainly situated in the northwest of the central area. The existing CMPBs are not closely related to these high-VVVs points and are often found in low-VVVs areas. Although Qingpu District has more high-VVVs points than Fengxian District, these points are relatively few and mostly concentrated in the northeastern region. There is no significant pattern in the distribution of existing CMPBs in Qingpu District.

Overall, the potential VVVs prediction analysis for CMPBs in the four administrative districts can assist designers in better understanding the current situation and conducting more informed planning. By considering relevant legal provisions and land-use restrictions [37], designers can select more suitable sites for CMPBs construction and work on improving the VVVs of existing CMPBs. We can provide effective guidance for planning operations from multiple aspects, such as site selection, planning strategies, scientific decision-making, resource allocation, and land-use efficiency based on the above model prediction results.

For a start, the potential area for site selection can be further clarified. Visualization of the prediction results of the ANN model for the four administrative districts in Shanghai using ArcGIS software (version 10.8.2) clearly shows the distribution of VVVs in different predicted areas. The prediction results show that the existing CMPBs are mainly distributed in the high-value areas of VVVs predicted by the model, and a few CMPBs are distributed in the low-value areas of VVVs predicted by the model. By comparing the raw data with the predicted data, the effectiveness of the ANN model training was verified, and other high-VVVs regions were also identified as potential sites for CMPBs. Planners can leverage these insights to focus their efforts on these high-potential areas. By prioritizing site selection for new CMPBs in these regions, planners can enhance the likelihood of successful and effective site placements, ultimately improving the overall vitality of the district.

Next, we can provide targeted guidance for differentiated planning and construction in designated areas. The distribution of high-VVVs points for CMPBs in Jiading District is relatively balanced, and the existing CMPBs are mostly located within the influence range of high-VVVs points. In the planning process, this characteristic can be addressed by further optimizing the service facilities around the existing CMPBs in the area, improving their service quality and attractiveness, and consolidating their vitality advantages. At the same time, in areas where high VVVs are located but CMPBs have not yet been constructed, targeted new museum planning can be carried out, focusing on coordinated development with surrounding service facilities that are evenly distributed, and creating a distinctive cultural area. At the same time, planning can be improved for weak links, such as Fengxian District, where high VVVs occur less frequently and are unevenly distributed. The model results reveal the problems in the correlation between infrastructure and CMPBs vitality. Planners can focus on strengthening the infrastructure construction planning for these areas based on this information, especially in low-VVVs areas, by reasonably arranging transportation facilities, commercial service facilities, etc., improving the regional environment, enhancing its attractiveness to CMPBs, and gradually improving the overall vitality level. For example, near the high-VVV point in the northwest direction of the Fengxian District central area, investment in cultural tourism-related service facilities can be increased to guide the construction and development of CMPBs.

Third, it can provide data support and comprehensively consider multiple factors for planning and layout. The model prediction results are presented in a data-driven and visualized manner. In the planning and decision-making process, subjective judgments or experience are no longer relied upon but can be analyzed based on the quantitative VVV data provided by the model and the relationship between infrastructure and vitality in various regions. At the same time, the ANN model comprehensively considers the impact of 13 types of secondary facility POI data on VVVs, which enables planners to comprehensively consider the synergistic effects of various infrastructure factors when planning CMPBs. This avoids the one-sidedness of a single factor dominating planning decisions, making the planning layout more scientific and reasonable. For example, when planning the functional zoning around CMPBs, the location and scale of commercial service facilities, educational service facilities, leisure and entertainment facilities, etc., should be reasonably arranged based on the impact weight of different types of facilities on vitality, forming a mutually reinforcing and organically integrated spatial layout to meet potential passenger flow needs and comprehensively enhance the attractiveness and vitality of the overall area.

In addition, planners can use quantitative data to promote the optimal allocation of regional resources. By predicting the distribution of VVVs through the model, planners can better understand the degree of demand for public resources in different regions. For high-VVVs prediction areas, more cultural, tourism, transportation, and other resources can be allocated reasonably to meet their development needs and further enhance vitality; For areas with low VVVs, resource shortcomings can be analyzed and targeted resource supplementation and improvement carried out. For example, in terms of transportation resource allocation, based on model prediction results, public transportation routes in high-VVVs areas are optimized and encrypted to improve tourist accessibility; At the same time, planning new transportation routes or improving existing transportation conditions in low-VVVs areas can attract more tourists and promote balanced regional development. Meanwhile, the predicted results of the model help planners accurately assess the potential for constructing CMPBs on different land parcels, thereby improving land-use efficiency. To avoid blind construction in low-vitality areas and minimize resource waste, limited land resources should be prioritized for the development and construction of high-vitality potential areas.

3.1.2. Small-Scale Regional Site Selection Prediction

To better address the needs of specific practical projects, we selected two CMPBs design practice projects currently being planned in Shanghai. We applied the same experimental process of grid cutting, data processing, and visualization analysis as in the large-scale macro prediction. Existing studies have shown that factors such as land use scope and distance scale affect project site selection [38,39].

However, for these projects, we significantly reduced the grid scale to obtain more accurate prediction results because the land use scope of the two projects is not suitable for large-scale grid cutting. We experimented with four grid scales: 5 M × 5 M, 10 M × 10 M, 20 M × 20 M, and 50 M × 50 M. Although the 5 M × 5 M and 10 M × 10 M grids provided higher precision, the experiments took significantly more time. The 50 M × 50 M grid, while faster, lacked sufficient accuracy. The 20 M × 20 M grid scale emerged as the optimal choice, balancing accuracy and efficiency. Therefore, we used a 20 M × 20 M grid scale for the grid cutting (Figure 12).

The first project is the Shanghai Industrial Museum, planned for construction in the World Expo Park in Huangpu District. According to the “Detailed Control Plan for the Cultural Expo Area of the Shanghai World Expo”, the 102-14 plot, with a land area of 33,335 square meters, is designated for this project. The surrounding land uses are varied and complex, including commercial service land, business office land, three types of residential cluster land, and early childhood land, all of which significantly influence the museum’s vitality [40,41]. According to our prediction results for urban development reserve land in the region, the VVVs in this area are “high in the west and low in the east”, with the high VVVs concentrated in the west and low VVVs in the east. The proposed plot for the Shanghai Industrial Museum falls within the MEDIUM–LOW-VVVs range (Figure 13).

In summary, in the future, we believe that the sustainable development of the Shanghai Industrial Museum should revolve around the following four aspects:

In the first place, optimizing spatial layout and functional settings. The ANN model predicts that the VVVs in the area where the museum is located will show a trend of “high in the west and low in the east”. Based on this prediction, in order to better develop, planners can actively guide the design of relevant shapes in the process of architectural design or landscape planning bidding. At the same time, in the later planning and operation, the main exhibition space, popular exhibition themes, and interactive experience areas with stronger attraction to visitors of the museum can be prioritized in the high-vitality area on the west side, fully utilizing the advantages of the high-VVVs area, attracting more tourists and extending their stay time, for example, setting up a large industrial equipment exhibition area and an immersive industrial history experience area in the west to enhance tourists’ participation and interest.

Targeted planning of museum ancillary facilities. For ancillary facilities such as dining and rest areas, souvenir shops, etc., the layout can be optimized based on the predicted tourist flow distribution of the model. Adding more rest seats and dining stalls or setting up distinctive souvenir shops in the high-VVVs area to meet the needs of tourists during their visit, while increasing consumption opportunities and enhancing the overall revenue of the museum. In the eastern area of low VVVs, the spatial attribute is “quiet space”, which allows for the planning of functional spaces that require quietness and learning, such as professional lecture halls, small research rooms, etc., to meet the usage needs of specific groups without deliberately increasing the spatial vitality of the area.

Thirdly, strengthen interaction and collaborative development with the surrounding environment. By taking advantage of the nature of the land around the museum, cooperation with surrounding commercial service land can be strengthened, and joint marketing activities can be carried out, such as collaborating with surrounding shopping malls to launch joint discount packages for shopping and visiting museums, attracting more consumers to come. In addition, cultural exchange and cooperation relationships have been established with commercial office land to provide services such as team visits and cultural training for enterprises, expand customer markets, and further stimulate the sustainable development potential of museums.

As a final point, improve surrounding transportation and connectivity. Considering that the overall VVVs of the proposed plot are of the MEDIUM–LOW level, emphasis can be placed on improving connectivity with surrounding transportation. Collaboration with the government’s transportation department can occur to optimize public transportation routes, such as adding bus stops to improve the convenience of public transportation. If conditions permit, add convenient connecting channels, such as pedestrian walkways, bike lanes, or shuttle bus routes to enhance the accessibility and experience of tourists, thereby increasing tourist flow.

The second project is the North Museum of the Shanghai Museum, planned for the riverside area of Yangpu District. This archaeological theme museum, centered around the Yangtze River Estuary No. 2 Ancient Ship, will use the 03G2-07 plot, which covers about 24,500 square meters. The surrounding land use is simpler, primarily consisting of public green space and commercial office land, which have a minimal impact on the museum’s vitality. Our research shows a similar “high in the west and low in the east” trend for VVVs in this area, with high VVVs concentrated in the northwest. Unlike the Shanghai Industrial Museum, the proposed plot for the North Museum of the Shanghai Museum has MEDIUM–HIGH VVVs, indicating a better overall performance (Figure 14).

We believe that the sustainable development of the North Branch of Shanghai Museum should revolve around the following three aspects: In the first place, an exhibition strategy based on location and characteristics. In the future, the North Branch of Shanghai Museum will strengthen the construction of the core themed exhibitions. The museum’s archaeological theme centered around the ancient ship No. 2 at the mouth of the Yangtze River is unique, and the model predicts that its VVVs will perform well overall and be “high in the west and low in the east”, with higher vitality in the northwest direction. Therefore, resources should be concentrated in the northwest direction to create a core archaeological themed exhibition area, showcasing archaeological achievements, cultural relic restoration processes, historical and cultural backgrounds related to the Yangtze River Estuary No. 2 ancient ship, and forming the core attraction of the museum. At the same time, a series of related themed exhibitions and activities should be planned around the archaeological theme in other areas of the museum. Based on the vitality of different areas, these exhibitions and activities should be reasonably distributed, and high-vitality areas should be used to display the most attractive core content. Low-vitality areas should be gradually activated through characteristic exhibitions and activities to enhance the visiting vitality of the entire museum.

Furthermore, strengthen integration and resource sharing with the surrounding environment. The museum is located in the riverside area of Yangpu District, surrounded mainly by public green spaces and commercial office land. Combining the concept of a 15 min living circle with the concept of sustainable development, the North Branch of Shanghai Museum can deeply integrate with the surrounding environment in the future to create a waterfront cultural landscape belt. Set up outdoor exhibition spaces, cultural and leisure squares, etc. in the riverside area, hold various cultural activities, attract more tourists, and provide leisure and cultural venues for surrounding business and office people, enhance the cultural atmosphere of the area, and gradually drive the low-VVVs area in the east. At the same time, actively promote the construction of shared surrounding public resources and facilities, and jointly plan leisure trails, bike lanes, etc., with surrounding public green spaces to achieve resource sharing. Tourists can enjoy the natural scenery of the riverside green space while visiting the museum, which increases the fun of the tour and enhances the overall visiting experience energy of the environment.

Ultimately, experience design is guided by quality and diversity. Planners optimize the internal spatial layout and visitor flow design of the museum based on the predicted distribution of VVVs using the model. Set up spacious and bright display spaces, convenient signage, and comfortable rest areas in high-vitality areas to ensure that visitors can smoothly explore various exhibition areas during their visit, avoiding congestion and inconvenience. In low-vitality areas, refer to the renovation method of the Shanghai Industrial Museum to improve space utilization. Overall, based on tourist feedback and behavior data (combined with model prediction results), continuously optimize service content and project experience to improve tourist satisfaction and loyalty. Through the development of strategies based on the model prediction results mentioned above, the Shanghai Industrial Museum and the North Branch of the Shanghai Museum can better leverage their advantages, develop in synergy with the surrounding environment, and achieve sustainable development.

In summary, we employed the ANN model to predict the VVVs of CMPBs using large-scale macro prediction and small-scale specific prediction. This process involved considering 13 types of Second Category service facilities, grid scale, and other factors. Judging from the experimental results, this part of the research can provide a reference for the planning and design of CMPBs in the city.

3.2. GA-Assisted CMPBs Improvement Service Facilities Environment

Genetic algorithms (GAs) are optimization methods inspired by the biological evolution process in nature. By simulating natural selection, crossover, and mutation, these algorithms seek the optimal or near-optimal solution from a problem’s solution space [42,43]. To address the challenge of low vitality in the site selection of CMPBs and to enhance the vitality of existing underdeveloped sites, we combined a trained ANN model with a GA. The fitness function employs the ANN model to predict output values, allowing the GA to assess each potential solution’s effectiveness in every iteration. Upon identifying an optimal solution, the GA uses the ANN model to predict the corresponding output once again. This approach effectively harnesses the predictive power of the ANN and the global optimization capability of the GA, thereby improving both efficiency and accuracy in problem-solving (Figure 15). Our combined algorithm aims to discover design strategies that enhance vitality by adjusting the number of public service facilities.

To ensure efficiency and accuracy, we randomly selected 120 objects from the original data set and organized the experimental groups according to the classification of urban service facilities in the POI data classification table (Table 2). The genetic calculations primarily involved adjusting the input values, with service facilities as inputs and VVVs as outputs. We modified the input values for each CMPB in various ways, measuring these changes and their impact on the output. “Changing the input value” involved setting new values for the number of service facilities and using a genetic algorithm to determine the optimal distribution of these values to maximize the VVVs of the CMPBs. Our experiments were categorized into three groups: “Second Category Service Facility Single Replacement”, “First Category Service Facility Classification Change”, and “Replace All Service Facilities”, resulting in a total of 90 experimental groups (Figure 16).

Second Category Service Facility Single Replacement. In this experiment, we replaced one of the 13 secondary service facilities in the data set while keeping the other 12 unchanged. The replacement facilities are assigned values of 500, 1000, 2000, 3500, and 5000, respectively.
First Category Service Facility Classification Change. In this experiment, we replaced one of the three primary service facilities in the data set, with the other two remaining unchanged. The value adjustments are based on the “Second Category Service Facility Single Replacement” experiment. For instance, “Government and Office Services” includes three secondary service facilities, resulting in five values: 1500, 3000, 6000, 10,500, and 15,000.
Replace All Service Facilities. This experiment involves replacing the data for all service facilities in the data set, assigning new values, and distributing them through genetic algorithms to create new data sets. Two schemes are employed in this experiment. Scheme A is based on the “Second Category Service Facility Single Replacement” approach, with values set at 6500, 13,000, 26,000, 45,500, and 65,000. Scheme B assigns values of 5000, 10,000, 20,000, 35,000, and 50,000, calculates the output values of the service facilities using genetic algorithms, and then distributes these values across 13 categories of secondary service facilities.

These experiments aim to understand how variations in the number and distribution of public service facilities impact the VVVs of CMPBs. The insights gained can enhance site selection and overall vitality. In these experiments, we defined nonlinear constraints to ensure that the inherited sum slightly exceeds both the original sum and the set value. For example, the value is set to 500, and the calculation process is (Equations (5)–(7)). The element corresponding to index

I

is selected from the global variable “fixed Inputs” and their sum calculated. The sum of the original selected values is defined as

S_{orig}

, the sum of all elements of the

\sum_{j = 1}^{n} x_{j}

variable

x

. At the same time, we combine the numerical authenticity affected by the distance of the three buffers. We define the genetic value to be strictly increasing at 500 M, 1000 M, and 2000 M, and the increase is at least 1. The calculation process is that

x_{1}

is 500 M,

x_{2}

is 1000 M, and

x_{3}

is 2000 M.

S_{orig} = \sum_{i \in I} f_{i}

(5)

\sum_{i \in I} f_{i} + 500 - \sum_{j = 1}^{n} x_{j} = 0 I

(6)

\{\begin{matrix} x_{1} - x_{2} + 1 \leq 0 \\ x_{2} - x_{3} + 1 \leq 0 \end{matrix}

(7)

3.2.1. Analysis of the Overall VVVs Improvement Ratio

We first analyzed and recorded the results of the new VVVs that exceeded the original values after the genetic experiment to determine if the experiment effectively enhanced the VVVs of the subjects. This analysis also aimed to address two key questions: First, whether VVVs consistently increase as the number of new service facilities grows. Second, how different types of service facilities affect VVVs differently. To better observe overall improvements, we converted the data into proportional values, indicating how much the new VVVs exceeded the original values, with 100% meaning that all new VVVs surpassed the original value. For the convenience of analysis and statistics, we define the original value as number 1 and the new VVVs as number 2.

As shown in Figure 17, the three types of experimental groups exhibited varying effectiveness in enhancing VVVs. Analyzing the incremental addition of service facilities, we observed that, overall, as the number of groups increased, the VVVs of most subjects improved. This suggests that increasing the number or coverage of service facilities positively impacts the vitality of CMPBs. For example, in the “Second Category Service Facility Single Replacement” experiment, the performance of Food Services in “GROUP_500” was 87.50%, improving progressively to 98.33% in “GROUP_3500.” On the other hand, Education Services had the lowest performance in “GROUP_500” among all 13 groups, at 76.67%, but it improved linearly, reaching 100% in “GROUP_5000”.

In the other two genetic experiments, an increase in the number of service facilities generally resulted in positive growth in the overall proportion. However, this growth did not continue indefinitely, as “growth saturation” and “growth extremes” occurred. For example, in the “Second Category Service Facility Single Replacement” experiment, after the number of Food Services increased to “GROUP_5000”, its value remained at 98.33%, identical to its performance in “GROUP_3500.” Similarly, Government Services showed no difference between the “GROUP_5000” and “GROUP_3500” groups, both at 99.17%. In the case of Tourism Services, the proportion in “GROUP_500” reached 100%, and despite further increases in the subsequent groups, its performance changed. These trends are observed in the other two categories as well.

These findings indicate that for service facilities that have reached saturation, further increasing the number or coverage may not significantly enhance service quality. Therefore, resources should be directed toward improving facilities that have not yet reached saturation to achieve a more balanced and efficient service configuration.

Additionally, we further analyzed the experimental results using the maximum value (MAX), minimum value (MIN), and average value (AVG), where MAX represents the upper limit of proportion improvement, MIN represents the lower limit, and AVG indicates the overall comprehensive improvement (Figure 18). We first evaluated the effectiveness of VVVs improvement across 90 experimental groups. As shown in the figure, all three metrics generally follow the trend of proportional increase with value growth.

Moreover, from the perspective of overall proportion improvement, the “First Category Service Facility Classification Change” experiment outperforms the “Second Category Service Facility Single Replacement” experiment, while the “Replace All Service Facilities” experiment shows the best overall improvement. This result suggests that, under ideal conditions, adding multiple service facilities in combination yields better performance than adding only one.

Among the 13 categories of secondary service facilities, Tourism Services demonstrated the best performance when the number of new facilities was the same. Additionally, facilities with a direct promotional relationship with CMPBs, such as Leisure Services, Traffic Services, Hotel Services, and Life Services, which provide transportation, accommodation, and daily services, also showed strong performance. Conversely, facilities with weaker relationships to CMPBs, such as Financial Services and Government Services, exhibited poorer performance.

3.2.2. Numerical Analysis of Overall VVVs Improvement

To evaluate the results of the genetic experiment with greater specificity, we employed two analytical methods. The integration of these methods provides deep insights into how different service facility configurations impact the vitality of CMPBs. Our study aims to address three key questions: First, how much value can VVVs gain by adding varying numbers of service facilities? Second, what impact do different types of service facilities have on VVVs when the same number of facilities are added? Third, does combining different types of service facilities or adding them individually yield the greatest improvement in VVVs, and what are the pros and cons of these two approaches?

The first method is standardized analysis, where we scale the original and new VVVs to a uniform range for balanced comparison and clearer trend visualization. This approach helps mitigate the impact of outliers and ensures a more accurate evaluation. In this normalization process, we scale both the original and new VVVs to a range of 0 to 1, enabling us to clearly compare the overall impact trends of different service facility configurations. Simultaneously, we analyze the absolute values to supplement our standardized results. By comparing absolute values, we aim to gain a comprehensive understanding of the data, considering subtle differences that may be obscured by standardization. This dual approach allows for a more nuanced evaluation of VVVs improvements under various parameters.

As shown in Figure 19 and Figure 20, three results can be obtained. First, as the number of new service facilities increases, the effect of various service facilities on increasing VVVs continues to increase. Second, when the number of service facilities increases by the same amount, Tourism Services has the best performance. This result shows that adding service facilities that have a stronger relationship with CMPBs can not only perform well in terms of the increase in proportion, but also obtain better results in terms of the increase in specific VVVs. Third, the study found that in experiments with a small number of new additions, the effect of the experimental group of different types of service facilities in improving VVVs is not significantly better than the experimental group of a single service facility. For example, in the “GROUP_500” experiment, the experimental group for Tourism Services performs better than the experimental group for Basic services.

However, as the number of new additions increases, the improvement effect of different types of service facilities combinations gradually becomes more outstanding. In addition, Scheme B performs better than Scheme A because Scheme B uses an algorithm combined with an ANN model to selectively allocate specific quantities to different service facilities, while Scheme A does not distribute the quantities evenly in a targeted manner. Therefore, it can be concluded that in the experiment of improving the VVVs of CMPBs, not only different quantities and different service facilities will affect the improvement effect, but how to effectively allocate different quantities to different service facilities in an overall plan will also have a huge impact on the improvement effect.

3.2.3. Analysis of the Promotion of VVVs of Songjiang District Museum

The Songjiang District Museum in Shanghai is situated north of Fangta Park within Songjiang District, Shanghai (Figure 21). Covering an area of 4700 square meters, with an exhibition hall spanning 1200 square meters, the museum prominently showcases cultural relics unearthed in the region and acquired for its collection8. According to the Songjiang District Master Plan and Land Use Master Plan (2017–2035) for Shanghai9, the area within 2000 m around the museum features various primary land uses, such as residential, industrial, parkland, and green spaces. To further analyze local land use, we consulted the Chinese Urban Basic Land Use Classification Dataset10 compiled by Gong et al. (2020). This comparison helped outline the comprehensive land use classification of Songjiang District, Shanghai.

Within a 500 m radius of the Songjiang District Museum, the primary land uses include residential areas, parks, educational and research facilities, and medical institutions. This area, ranging from 1000 to 2000 square meters, also contains industrial zones, administrative offices, commercial areas, and sports and cultural facilities. The diverse land use creates a supportive environment for the development of various service facilities, fostering a multifunctional service ecosystem that can sustainably enhance the museum’s service level.

To explore this, we plan to use algorithms to simulate how a new service environment, consisting of 13 categories of second-level service facilities, impacts visitor vitality at the Songjiang District Museum through three types of experiments. In addition to examining how different numbers and types of service facilities affect overall VVVs, we will address four specific issues:

Efficiency of Service Facility Types: Analyze which service facility types are most and least efficient at improving VVVs when adding either a small or large number of facilities.
Performance of Facility Combinations: Evaluate the effectiveness of various service facility combinations in enhancing VVVs and investigate the reasons behind the differences in their performance.
Allocation Strategy Analysis: Study Scheme B’s approach to allocating new service facilities and its implications.
Feasibility Analysis: Assess the practical effectiveness of different types of service facilities based on algorithmic validation in real-world scenarios.

As shown in Figure 22, in the “Second Category Service Facility Single Replacement” experiment, only the “GROUP_500” for Education Services failed to surpass the original VVVs. Among the experimental groups that varied by the number of facilities, Financial Services showed the least improvement. This is due to the more indirect role Financial Services play in the sustainable development of CMPBs.

In contrast, Hotel Services, Tourism Services, Life Services, and Traffic Services directly benefit CMPBs, resulting in better performance. Notably, Shopping Services did not significantly improve performance. This lack of effectiveness is attributed to the “competition and diversion” between Shopping Services and CMPBs in certain contexts. Conversely, Tourism Services and Leisure Services, sharing similar cultural and entertainment attributes, demonstrated a strong “cooperative growth” effect.

Two additional phenomena were observed in the experiments combining different service facilities. First, the relationship between the number of service facilities and the improvement in VVVs is not always linear. For instance, Tourism Services in the “GROUP_5000” experiment showed some fluctuations. Second, the experimental group for Government and Office Services experienced a significant “growth reversal,” where adding more facilities led to a decrease in VVVs. This negative effect was also observed in the “GROUP_5000” for Business and Economic Services and Scheme A.

The experimental groups discussed above represent artificial and average scenarios for service facility environments, allowing us to observe the effects of varying quantities of different service facilities. Scheme B, which has shown excellent performance in enhancing VVVs, operates strategically through our algorithm. Therefore, it is essential to analyze how Scheme B combines different types of service facilities.

Figure 23 and Figure 24 reveal that the allocation weights for various types of service facilities fluctuate across different experimental groups. For instance, in the “VALUE_5000” group, Traffic Services have the highest allocation weight, while in the “VALUE_10000” group, Company Services take the lead. However, Company Services have a lower allocation weight in the other four groups. Although Traffic Services does not always receive the highest allocation, it consistently receives a substantial portion. Special cases, such as “VALUE_10000”, indicate that the algorithm strategically adjusts weights, favoring specific service types to maximize visitor vitality. This idealized operation may not directly translate into real-world guidance, as the synergistic benefits of Company Services and CMPBs may not be as pronounced in practice. Implementing such an allocation in reality could lead to unbalanced urban development and inefficient service provision.

Additionally, the allocation weights for Food Services, Leisure Services, and others show significant fluctuations. Despite Scheme B’s significant improvements in VVVs, its direct application in actual construction may be limited and requires further human consideration. For practical applications, particularly in the case of the Songjiang Museum, it is essential to assess the feasibility of construction based on the results from other experimental groups. If 500 new service facilities are to be added, “Tourism Services” show the best performance, while Education Services provide the least benefit. Leisure Services and Traffic Services also demonstrate relatively good results. When combining different service facilities, Basic Services also perform well. However, the construction costs of various service facilities vary significantly in reality. For example, Tourism Services and Leisure Services are more expensive to build compared to Food Services and Financial Services. Therefore, project planning should consider construction costs to effectively promote the sustainable development of the Songjiang District Museum.

4. Conclusions and Discussion

In the context of transforming and developing high-quality urban areas, this study introduces new perspectives and tools for professionals in the planning, design, and vitality enhancement of CMPBs. We developed and validated the ANN model that connects the distribution of urban infrastructure with VVVs of CMPBs. By integrating the GA, the model explores new development strategies by altering infrastructure configurations. However, achieving high accuracy in training a neural network for predictive problems necessitates extensive data collection. This process is complex and time-consuming, requiring a team of experts to preliminarily evaluate the data’s validity, which limits the widespread application of ANNs.

Despite our extensive and pioneering work, we have found that the limitations of this study mainly lie in the following three aspects. Firstly, the limitations and incompleteness of data collection. The data collected by this research institute do not come from the government, but from enterprises, so there may be issues with incomplete coverage of regions, inclusion of groups, or untimely data updates. Another issue is that some niche or emerging service facilities may not be accurately included in POI data, resulting in an incomplete evaluation of the surrounding infrastructure by the ANN model we ultimately trained. Secondly, the regional nature of the data limits the performance of the model. This study only used Shanghai as an example for data collection and model training. Although Shanghai has a certain representativeness, there are differences in social structure, cultural economy, and people’s preferences among different cities, which means that models trained based on Shanghai data may not be directly applicable to CMPBs planning in other cities. This has also been confirmed in our subsequent research. Therefore, for example, in certain historical and cultural cities, there may be unique cultural tourism needs and infrastructure layout patterns, and the ANN model used for research may need to re-collect data, adjust training, and validate. Thirdly, there is a lack of human perception and subjective experiential factors. Our research, which focuses on objective factors such as public service facilities, has not fully considered the driving effects of human perception, architectural form, color composition, and other factors on tourist experience and visitor vitality. However, in practice, these factors may have a significant impact on tourists’ willingness to visit and their behavior.

In our case studies, we demonstrated the model’s application in planning and constructing CMPBs at various geographical scales. The experiments showed that different infrastructure combinations and scales affect the VVVs of CMPBs. For example, increasing traffic facilities and tourist attractions positively impacts VVVs, while adding corporate and real estate facilities has less effect. These findings contribute to understanding the interaction between CMPBs and infrastructure, aiding in better urban design at a macro scale [44]. We conducted 90 experiments combining the ANN model with the GA. The results indicated that the calculated new service facility environments effectively improved the VVVs of different CMPBs in the data set. A specific study on the Songjiang District Museum in Shanghai analyzed the impact of various scales and types of service facilities, such as “Tourism Services”, “Life Services”, and “Food Services”. The study emphasized considering economic and social factors in the actual development process and using algorithm-recommended service facility weights for sustainable development.

This study recognizes that CMPBs’ vitality is not solely influenced by infrastructure but considers basic service facilities as a significant factor [32]. While addressing the gap in vitality analysis and development for CMPBs site selection, the ANN model’s applicability varies by city due to differing infrastructure distributions and CMPBs characteristics. Testing in Shenzhen and Guangzhou revealed significant prediction errors when using a model trained on Shanghai data. Therefore, it is crucial to collect relevant data sets for different urban conditions to achieve accurate results.

Overall, this study has some shortcomings, but it has some enlightening significance for future research on related topics. Researchers can conduct breakthrough research in the following five aspects:

Multi-source data fusion and validation. For future research, we suggest focusing on collecting and integrating more diverse and authoritative data sources to improve the accuracy, comprehensiveness, and reliability of the data. In research, cross-validation and supplementation of data from different sources can be further combined with government statistical data, field research data, and research reports from professional institutions.
Targeted enhancement of multi-regional data collection and model adaptability research. Our suggestion is to conduct research on multiple cities, and that it is necessary to collect data from different types of cities and regions in order to further investigate the relationship patterns between CMPBs in multiple cities and the surrounding complex environment under different backgrounds. On this basis, comparative analysis of multi-source data from different regions is necessary to develop models with stronger adaptability and minimize the limitations of regional differences on model application. This will help to promote research results to a wider range of regions and provide more targeted guidance for CMPBs planning in various regions.
Combining cutting-edge technology to further enhance the model’s ability to handle extreme values and complex situations. All advanced algorithms and models can be continuously optimized, and our optimized ANN model and GA algorithm also need to be continuously updated by subsequent researchers. Subsequent researchers need to further explore new algorithms or improve existing model architectures to further enhance the model’s ability to identify and process extreme values and special situations. For example, techniques such as anomaly detection algorithms and Generative Adversarial Networks (GANs) can be further introduced to enable the model to better cope with various complex and ever-changing situations in reality.
Enhance the visual interpretability of the model. In this study, we used Python (version 3.11.5)code to draw results analysis graphs and ArcGIS software (version 10.8.2) for multi-dimensional visualization presentation. We suggest further research, if there is a stronger team, to continue enriching explanatory methods and developing visualization tools to help planners better understand the decision-making process and output results of the model. At the same time, there is another challenging task, which is a huge innovative work. We suggest that researchers explore the transformation of complex neural network models into more easily understandable rule sets or model simplification techniques, reduce the understanding threshold of the model, and further improve operability and convenience.
A research trend that considers multiple factors comprehensively. Future research can further consider the subjective perception influence of people in space and collect physiological and psychological response data of tourists in the museum through devices such as EEG instruments and skin conductance response instruments. Then, in the new research, we can analyze how these subjective factors interact with objective environmental factors. For example, we can study how architectural spatial layout, garden landscape, and color design affect tourists’ emotions and behaviors, thereby affecting visit duration and satisfaction. This will provide a more scientific basis for museum architectural design and spatial optimization and create a more attractive tourist environment. In addition, we strongly recommend that future research pays more attention to the dynamic changes and impacts of social and cultural trends, attaches importance to the study of changes in social and cultural trends, and timely incorporates factors such as trendy cultural phenomena, highly popular design styles, and changes in social and cultural orientations into the research framework. For example, time series analysis and other methods can be used to predict the long-term impact of social and cultural changes on the demand of CMPBs, adjust planning strategies in advance, and ensure that CMPBs maintain positive vitality and strong attraction in the changing complex cultural environment.

In the planning and vitality enhancement of CMPBs, sustainable development is a critical focus. Several key considerations should be addressed:

Synergistic Growth. The “synergistic growth” phenomenon should be leveraged by aligning service facilities with the land use attributes of the city. Service facilities with similar attributes or direct impacts can create synergistic effects. For instance, placing such facilities near commercial and service areas can boost CMPB visibility by attracting commercial traffic.
Competitive Diversion: A high concentration of similar service facilities, such as multiple Tourism or Leisure Services within the same area, may not enhance CMPBs’ sustainable development. Instead, it could lead to competition and fragmentation of visitor traffic, reducing overall effectiveness.
Growth Saturation: Each region and CMPB has a specific estimated value for VVVs, and there is a saturation threshold for regional carrying capacity. Over-reliance on one type of service facility or excessive construction of similar facilities may not continuously improve attractiveness. Instead, a diverse range of service facilities is needed to ensure balanced development.
Growth Reversal: The “growth reversal” phenomenon highlights that each region has a limited carrying capacity. While adding service facilities can initially enhance CMPB vitality, continuous increases can lead to diminishing returns. Rising construction, labor, and operational costs can contribute to “growth bottlenecks”, “growth fluctuations”, and eventually “reverse growth”, where the benefits of additional facilities decline. Strategic thinking and a balanced approach are essential throughout the construction process to address these issues effectively [45].

According to the official document released by the Shanghai Municipal People’s Government on the comprehensive implementation of the “Shanghai Urban Master Plan (2017–2035)”, Shanghai is striving to build an international cultural metropolis brand from multiple aspects. The core goal is to build an international cultural and creative industry center city and a world-renowned tourist destination city. Shanghai already has relatively abundant CMPBs resources, and it is an important issue to flexibly and efficiently utilize these resources. At the same time, Shanghai is vigorously building the “Five New Cities” in administrative districts such as Songjiang District, and there will be the construction of new CMPBs in the future. Although this study has some limitations, it can still effectively provide stronger support for the scientific planning and sustainable development of CMPBs to a certain extent. At the same time, the innovative interdisciplinary research method used in the study will also bring new ideas and methods to CMPBs planning research, improving the scientific and practical nature of the research.

Author Contributions

B.Z.: Research design, investigation, execution, data curation, software, formal analysis, validation, writing—original draft, visualization; H.Z.: conceptualization, methodology, software, resources, writing—review and editing, supervision, project administration, funding acquisition. X.C.: writing—review and editing, validation, supervision, project administration, investigation, conceptualization. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the New Faculty Start-up Grant (Project No. 9610653) from the City University of Hong Kong.

Data Availability Statement

The raw data supporting the conclusions of this article will be made available by the authors on request.

Acknowledgments

We sincerely thank the editors and reviewers for their valuable comments and suggestions.

Conflicts of Interest

The authors declare no conflicts of interest.

Notes

1	https://mori-m-foundation.or.jp/english/ius2/gpci2/ (accessed on 11 November 2023).
2	https://www.mohurd.gov.cn/gongkai/zhengce/zhengcefilelib/201507/20150703_224143.html (accessed on 6 November 2023).
3	https://lbs.amap.com/ (accessed on 2 December 2023)
4	https://lbs.amap.com/api/mobility-standard/beforeservice/destination/geoios (accessed on 2 December 2023)
5	https://lbsyun.baidu.com/ (accessed on 3 December 2023)
6	https://www.resdc.cn/DOI/DOI.aspx?DOIID=122 (accessed on 5 December 2023)
7	https://www.dianping.com/ (accessed on 8 December 2023)
8	https://www.songjiang.gov.cn/zjsj/stepin.html#page2 (accessed on 4 February 2024)
9	https://hd.ghzyj.sh.gov.cn/ghsp/ghsp/shj/201905/t20190516_915877.html (accessed on 15 February 2024)
10	https://data-starcloud.pcl.ac.cn/zh/resource/7 (accessed on 19 January 2024)

References

Yao, J.; Zhang, X.X.; Murray, A.T. Location optimization of urban fire stations: Access and service coverage. Comput. Environ. Urban Syst. 2019, 73, 184–190. [Google Scholar] [CrossRef]
Sakai, T.; Beziat, A.; Heitz, A. Location factors for logistics facilities: Location choice modeling considering activity categories. J. Transp. Geogr. 2020, 85, 10. [Google Scholar] [CrossRef]
Feng, H.; Tang, X.; Zou, C. Optimizing the Layout of Service Facilities for Older People Based on POI Data and Machine Learning: Guangzhou City as an Example. Land 2024, 13, 700. [Google Scholar] [CrossRef]
Tang, X.; Zou, C.; Shu, C.; Zhang, M.; Feng, H. Research on Site Selection Planning of Urban Parks Based on POI and Machine Learning—Taking Guangzhou City as an Example. Land 2024, 13, 1362. [Google Scholar] [CrossRef]
JGJ66-2015; Design Code for Museum Buildings. China Architecture&Building Press: Beijing, China, 2015.
Li, Y.; Huang, J.X.; Yang, L.C. From Functional Space to Experience Space: Applying space syntax analysis to a museum in China. Int. Rev. Spat. Plan. Sustain. Dev. 2020, 8, 86–99. [Google Scholar] [CrossRef]
Wang, Z.W.; Wang, X.C.; Dong, Z.J.; Li, L.S.; Li, W.J.; Li, S.C. More Urban Elderly Care Facilities Should Be Placed in Densely Populated Areas for an Aging Wuhan of China. Land 2023, 12, 220. [Google Scholar] [CrossRef]
Zhang, X. Study on the Preparation of Shanghai Gardensand Landscape Architecture Museum. China Acad. J. Electron. Publ. House 2021, 37, 127–131. [Google Scholar] [CrossRef]
Sun, H. The Characteristies of Heritage Site Museums and Their Planning and Constructions. Southeast Cult. 2022, 14–24+191–192. [Google Scholar] [CrossRef]
Quan, D.J.; Li, J.W.; Liu, X.C. Analysis on location of Rock Art Museum in Helan Mountain. Hum. Geogr. 2004, 19, 67–70. [Google Scholar]
Wei, Y.X.; Zhang, X.X.; Shi, Y.; Xia, L.; Pan, S.; Wu, J.S.; Han, M.J.; Zhao, X.Y. A review of data-driven approaches for prediction and classification of building energy consumption. Renew. Sust. Energ. Rev. 2018, 82, 1027–1047. [Google Scholar] [CrossRef]
Zhang, G.Q.; Patuwo, B.E.; Hu, M.Y. Forecasting with artificial neural networks: The state of the art. Int. J. Forecast. 1998, 14, 35–62. [Google Scholar] [CrossRef]
Marzban, C. Basic statistics and basic AI: Neural networks. In Artificial Intelligence Methods in the Environmental Sciences; Springer: Dordrecht, The Netherlands, 2009; pp. 15–47. [Google Scholar]
Pijanowski, B.C.; Tayyebi, A.; Doucette, J.; Pekin, B.K.; Braun, D.; Plourde, J. A big data urban growth simulation at a national scale: Configuring the GIS and neural network based Land Transformation Model to run in a High Performance Computing (HPC) environment. Environ. Modell. Softw. 2014, 51, 250–268. [Google Scholar] [CrossRef]
Grekousis, G.; Manetos, P.; Photis, Y.N. Modeling urban evolution using neural networks, fuzzy logic and GIS: The case of the Athens metropolitan area. Cities 2013, 30, 193–203. [Google Scholar] [CrossRef]
Tsagkis, P.; Bakogiannis, E.; Nikitas, A. Analysing urban growth using machine learning and open data: An artificial neural network modelled case study of five Greek cities. Sustain. Cities Soc. 2023, 89, 104337. [Google Scholar] [CrossRef]
Zhou, G.Y.; Zhu, Z.W.; Luo, S.M. Location optimization of electric vehicle charging stations: Based on cost model and genetic algorithm. Energy 2022, 247, 11. [Google Scholar] [CrossRef]
Alogdianakis, F.; Dimitriou, L.; Charmpis, D.C. Data-driven recognition and modelling of deterioration patterns in the US National Bridge Inventory: A genetic algorithm-artificial neural network framework. Adv. Eng. Softw. 2022, 171, 103148. [Google Scholar] [CrossRef]
Hajimohammadi Tabriz, Z.; Taheri, M.H.; Khani, L.; Çağlar, B.; Mohammadpourfard, M. Enhancing a bio-waste driven polygeneration system through artificial neural networks and multi-objective genetic algorithm: Assessment and optimization. Int. J. Hydrogen Energy 2024, 58, 1486–1503. [Google Scholar] [CrossRef]
Taghipour Kani, G.; Ghahremani, A. Optimal design of heat pipes for city gate station heaters by applying genetic and Bayesian optimization algorithms to an artificial neural network model. Case Stud. Therm. Eng. 2024, 55, 104203. [Google Scholar] [CrossRef]
Saryazdi, S.m.E.; Etemad, A.; Shafaat, A.; Bahman, A.M. Data-driven performance analysis of a residential building applying artificial neural network (ANN) and multi-objective genetic algorithm (GA). Build. Environ. 2022, 225, 109633. [Google Scholar] [CrossRef]
Barros, F.S.; Graça, P.A.; Lima, J.J.G.; Pinto, R.F.; Restivo, A.; Villa, M. Using Recurrent Neural Networks to improve initial conditions for a solar wind forecasting model. Eng. Appl. Artif. Intell. 2024, 133, 108266. [Google Scholar] [CrossRef]
He, J.; Zheng, H. Prediction of crime rate in urban neighborhoods based on machine learning. Eng. Appl. Artif. Intell. 2021, 106, 104460. [Google Scholar] [CrossRef]
Sundermeyer, M.; Ney, H.; Schlüter, R. From Feedforward to Recurrent LSTM Neural Networks for Language Modeling. IEEE-ACM Trans. Audio Speech Lang. 2015, 23, 517–529. [Google Scholar] [CrossRef]
Gu, J.X.; Wang, Z.H.; Kuen, J.; Ma, L.Y.; Shahroudy, A.; Shuai, B.; Liu, T.; Wang, X.X.; Wang, G.; Cai, J.F.; et al. Recent advances in convolutional neural networks. Pattern Recognit. 2018, 77, 354–377. [Google Scholar] [CrossRef]
Cheng, E.W.L.; Li, H. Exploring quantitative methods for project location selection. Build. Environ. 2004, 39, 1467–1476. [Google Scholar] [CrossRef]
Cheng, E.W.L.; Li, H.; Yu, L. A GIS approach to shopping mall location selection. Build. Environ. 2007, 42, 884–892. [Google Scholar] [CrossRef]
Gulshad, K.; Szydłowski, M.; Yaseen, A.; Aslam, R.W. A comparative analysis of methods and tools for low impact development (LID) site selection. J. Environ. Manag. 2024, 354, 120212. [Google Scholar] [CrossRef]
Qian, J.; Liu, Z.; Du, Y.; Liang, F.; Yi, J.; Ma, T.; Pei, T. Quantify city-level dynamic functions across China using social media and POIs data. Computers Environ. Urban Syst. 2021, 85, 101552. [Google Scholar] [CrossRef]
Ma, C.-X.; Peng, F.-L. Evaluation of spatial performance and supply-demand ratios of urban underground space based on POI data: A case study of Shanghai. Tunn. Undergr. Space Technol. 2023, 131, 104775. [Google Scholar] [CrossRef]
Shu, Z.W.; Carrasco, R.A.; Sánchez-Montañés, M.; García-Miguel, J.P. A Multi-Criteria Decision Support Model for Restaurant Selection Based on Users’ Demand Level: The Case of Dianping.com. Inf. Process. Manager 2024, 61, 24. [Google Scholar] [CrossRef]
Song, L.; Kong, X.S.; Cheng, P. Supply-demand matching assessment of the public service facilities in 15-minute community life circle based on residents’ behaviors. Cities 2024, 144, 104637. [Google Scholar] [CrossRef]
Elias, R.; Issa, R.R.A. Artificial Neural Network-Based Generative Design Optimization of the Energy Performance of Florida Single-Family Houses. J. Comput. Civil. Eng. 2024, 38, 13. [Google Scholar] [CrossRef]
Zheng, H.; Yuan, P.F. A generative architectural and urban design method through artificial neural networks. Build. Environ. 2021, 205, 16. [Google Scholar] [CrossRef]
Grekousis, G. Artificial neural networks and deep learning in urban geography: A systematic review and meta-analysis. Computers, Environ. Urban Syst. 2019, 74, 244–256. [Google Scholar] [CrossRef]
Zhang, L. Evaluating the Effects of Size and Precision of Training Data on ANN Training Performance for the Prediction of Chaotic Time Series Patterns. Int. J. Softw. Sci. Comput. Intell. 2019, 11, 16–30. [Google Scholar] [CrossRef]
Zhu, H.; Gao, X.; Tian, W.; Wei, M.; Zhang, X.; Chen, Y. Investigating the mechanisms of sustainable urban land renewal: A perspective of spatial development rights. Sustain. Cities Soc. 2024, 109, 105536. [Google Scholar] [CrossRef]
Jeong, J.S.; García-Moruno, L.; Hernández-Blanco, J. A site planning approach for rural buildings into a landscape using a spatial multi-criteria decision analysis methodology. Land Use Policy 2013, 32, 108–118. [Google Scholar] [CrossRef]
Bilgili, A.; Arda, T.; Kilic, B. Explainability in wind farm planning: A machine learning framework for automatic site selection of wind farms. Energy Conv. Manag. 2024, 309, 118441. [Google Scholar] [CrossRef]
He, S.; Zhang, Z.; Yu, S.; Xia, C.; Tung, C.-L. Investigating the effects of urban morphology on vitality of community life circles using machine learning and geospatial approaches. Appl. Geogr. 2024, 167, 103287. [Google Scholar] [CrossRef]
Wu, W.; Dang, Y.; Zhao, K.; Chen, Z.; Niu, X. Spatial nonlinear effects of urban vitality under the constraints of development intensity and functional diversity. Alex. Eng. J. 2023, 77, 645–656. [Google Scholar] [CrossRef]
Feng, L.; Mi, X.; Yuan, D. Optimal planning of urban greening system in response to urban microenvironments in a high-density city using genetic algorithm: A case study of Tianjin. Sustain. Cities Soc. 2022, 87, 104244. [Google Scholar] [CrossRef]
Li, J.; Liu, Z.; Wang, X. Public charging station location determination for electric ride-hailing vehicles based on an improved genetic algorithm. Sustain. Cities Soc. 2021, 74, 103181. [Google Scholar] [CrossRef]
Khavarian-Garmsir, A.R.; Sharifi, A.; Sadeghi, A. The 15-minute city: Urban planning and design efforts toward creating sustainable neighborhoods. Cities 2023, 132, 104101. [Google Scholar] [CrossRef]
Ge, Y.; Jia, W.; Zhao, H.; Xiang, P. A framework for urban resilience measurement and enhancement strategies: A case study in Qingdao, China. J. Environ. Manag. 2024, 367, 122047. [Google Scholar] [CrossRef] [PubMed]

Figure 1. Study area. (a) Shanghai’s location in China; (b) Distribution of Shanghai’s 16 districts.

Figure 2. Research framework.

Figure 3. Distribution of POI data of 13 types of second category service facilities in Shanghai.

Figure 4. Total of 344 CMPBs that meet the criteria. (a,b) They are CMPBs-POI, CMPBs nuclear density analysis.

Figure 5. Three levels of buffer zones of CMPBs. (a–c) 500 M buffer, 1000 M buffer, 2000 M buffer.

Figure 6. Matrix heat map of 13 types of Second Category service facilities and VVVs in the three levels of buffer zones of CMPBs. The significance level in the figure is represented as * p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001, and ns p > 0.05.

Figure 7. Specific conditions of 13 types of Second Category service facilities and VVVs in the three levels of buffer zones of CMPBs. (a–m) Food services, Residential Services, Company Services, Shopping Services, Traffic Services, Financial Services, Hotel Services, Tourism Services, Life Services, Leisure Services, Education Services, Hospital Services, Government Services.

Figure 8. ANN model for research settings.

Figure 9. Training epoch and loss.

Figure 10. Comparison of ANN model training with other models.

Figure 11. VVVs prediction results of four districts in Shanghai, (a–d): Fengxian, Jiading, Qingpu, Songjiang.

Figure 12. Comparison of scene grid scales. (a) The grid scale of the Shanghai Industrial Museum; (b) The grid scale of the Shanghai Museum North Building; (0–5) are SHP Plot File, Grid Overlay Analysis, Grid Scale: 5 m × 5 m, 10 m × 10 m, 20 m × 20 m, 50 m × 50 m.

Figure 13. Shanghai Industrial Museum site situation and VVVs prediction results. (a–c) are site planning situation, research and design scope map, and design scope vitality value prediction. (a,b) the pictures provided by Shanghai Urban Planning and Design Institute were redrawn.

Figure 14. Shanghai Museum North Building site situation and VVVs prediction results. (a–d) are site planning situation, research and design scope map, specific construction land schematic, design scope vitality value prediction. (a–c) are redrawn from the pictures provided by the Shanghai Museum.

Figure 15. The technical route of combining a genetic algorithm and an ANN model.

Figure 16. Grouping of genetic algorithm experiments (90 groups).

Figure 17. The proportion of new VVVs greater than the original value. (a) “Second Category Service Facility Single Replacement” experiment. (b) “First Category Service Facility Classification Change” experiment. (c) “Replace All Service Facilities” experiment.

Figure 18. MAX, AVG, MIN statistics of the proportion of new VVVs greater than the original value.

Figure 19. Normalization analysis of all experimental groups except Scheme B.

Figure 20. Box plot analysis of all experimental groups.

Figure 21. The current situation of Songjiang Museum. (a) Satellite images of the surrounding environment. (b) Classification of land use.

Figure 22. Numerical analysis of VVVs improvement in museums in Songjiang. (a) Comparing all experimental groups except Scheme B. (b) Comparison of “First Category Service Facility Classification Change” experiment. (c) Comparing the “Replace All Service Facilities” experiment.

Figure 23. The total number of facilities allocated by Scheme B for different types of service facilities at different distances.

Figure 24. Statistics on the allocation ratio of new service facilities of different types with different quantities and distances in Scheme B.

Table 1. CMPBs classification table.

Classification	Category	Content
Collection Nature and Display Content	Art Category	Mainly display the artistic and aesthetic value of the collection, such as an art museum.
	History Category	Display collections from a historical perspective, such as history museums and memorial halls.
	Science and Technology Category	Display the natural world in a classification, development or ecological way, and use a three-dimensional method to show scientific results from macro or micro aspects. Such as natural history museum, technology museum and science and technology museum.
	Comprehensive category	Comprehensive display of local nature, history, revolutionary history, art collections, etc.

Table 2. POI data classification of urban public service facilities.

First Category	Second Category	Number of POIs
Basic Services	Traffic Services	111,561
	Education Services	40,999
	Leisure Services	32,705
	Tourism Services	6760
	Life Services	168,359
	Hospital Services	23,135
Business and Economic Services	Financial Services	13,116
	Hotel Services	7728
	Shopping Services	199,661
	Food Services	142,880
Government and Office Services	Government Services	52,315
	Company Services	145,266
	Residential Services	41,989

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

© 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

Share and Cite

MDPI and ACS Style

Zhao, B.; Zheng, H.; Cheng, X. A Machine Learning Approach to Predict Site Selection from the Perspective of Vitality Improvement. Land 2024, 13, 2113. https://doi.org/10.3390/land13122113

AMA Style

Zhao B, Zheng H, Cheng X. A Machine Learning Approach to Predict Site Selection from the Perspective of Vitality Improvement. Land. 2024; 13(12):2113. https://doi.org/10.3390/land13122113

Chicago/Turabian Style

Zhao, Bin, Hao Zheng, and Xuesong Cheng. 2024. "A Machine Learning Approach to Predict Site Selection from the Perspective of Vitality Improvement" Land 13, no. 12: 2113. https://doi.org/10.3390/land13122113

APA Style

Zhao, B., Zheng, H., & Cheng, X. (2024). A Machine Learning Approach to Predict Site Selection from the Perspective of Vitality Improvement. Land, 13(12), 2113. https://doi.org/10.3390/land13122113

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Menu