CN118586643A - Service-oriented manufacturing resource optimization scheduling system based on adaptive learning algorithm - Google Patents

Abstract

The present invention discloses a service-oriented manufacturing resource optimization and scheduling system based on an adaptive learning algorithm, which relates to the technical field of scheduling services, including a data acquisition and preprocessing module, a data analysis and feature extraction module, an adaptive learning algorithm training module, a multi-objective optimization scheduling module, and a real-time execution and feedback adjustment module. The service-oriented manufacturing resource optimization and scheduling system based on an adaptive learning algorithm, by introducing an adaptive learning algorithm, enables the system to learn and optimize resource scheduling strategies in real time, thereby improving the efficiency of resource scheduling, and enhances the real-time adjustment capability of the system, so that the system can adjust the resource scheduling strategies in real time according to changes in the production environment to cope with various emergencies or changes in demand, and at the same time, improves the prediction capability, by analyzing and learning historical data, enables the system to predict future production needs, thereby making resource scheduling preparations in advance, and making full use of data resources to improve resource utilization and production efficiency.

Description

Service-oriented manufacturing resource optimization scheduling system based on adaptive learning algorithm

Technical Field

The present invention relates to the technical field of scheduling services, and in particular to a service-oriented manufacturing resource optimization scheduling system based on an adaptive learning algorithm.

Background Art

Service-oriented manufacturing, as an important model of modern manufacturing, emphasizes the integration of service elements into the production process to meet the diverse and personalized needs of customers. In this model, the manufacturing process is no longer a simple product processing and assembly, but a complex system that integrates a series of service activities such as design, production, sales, and maintenance. Therefore, how to efficiently and flexibly dispatch and manage various resources has become a problem that needs to be solved in the field of service-oriented manufacturing.

Traditional resource scheduling methods, such as rule-based methods and heuristic algorithms, can achieve good results in some cases, but they often lack flexibility and adaptability. On the one hand, these methods usually rely on fixed rules and parameters, which are difficult to cope with the ever-changing needs and conditions in the production environment; on the other hand, they can only handle simple and static problems, and cannot effectively deal with complex, changeable and dynamically changing actual situations.

In recent years, with the rapid development of artificial intelligence and machine learning technologies, adaptive learning algorithms have been widely used in various fields. These algorithms can gradually adapt to changes in the environment and find the optimal solution through continuous learning and optimization. Therefore, applying adaptive learning algorithms to the field of service-oriented manufacturing resource scheduling has great potential and advantages.

At present, there are studies on service-oriented manufacturing at home and abroad, but the research on service-oriented manufacturing resource optimization scheduling system based on adaptive learning algorithm is still blank, resulting in the following deficiencies in the existing scheduling service of service-oriented manufacturing resources:

First, existing technologies currently lack predictive capabilities and are temporarily unable to predict and judge future production needs, which results in a lack of foresight in resource scheduling and an inability to effectively respond to future changes.

Secondly, existing technologies usually rely on fixed scheduling rules or heuristic algorithms, which may no longer be applicable in a changing production environment, resulting in unreasonable resource allocation and low production efficiency. In addition, traditional resource scheduling systems often do not make full use of the data generated during the production process and are unable to use the rich information contained in these data to optimize resource scheduling strategies.

Therefore, it is urgent to improve this shortcoming. The present invention studies and improves the existing structure and deficiencies, and provides a service-oriented manufacturing resource optimization scheduling system based on an adaptive learning algorithm.

Summary of the invention

The purpose of the present invention is to provide a service-oriented manufacturing resource optimization scheduling system based on an adaptive learning algorithm to solve the problems raised in the above background technology.

To achieve the above-mentioned purpose, the present invention provides the following technical solution: a service-oriented manufacturing resource optimization scheduling system based on an adaptive learning algorithm, comprising:

A data acquisition and preprocessing module, which is used to collect raw data and perform preprocessing operations on the data;

A data analysis and feature extraction module, which uses data analysis technology to extract key features from the data for training the adaptive learning model;

An adaptive learning algorithm training module, which uses deep learning and reinforcement learning algorithms, combined with historical data and real-time feedback, to continuously train and optimize resource scheduling models;

A multi-objective optimization scheduling module, which generates multiple resource scheduling schemes based on the output of the adaptive learning algorithm, taking into account multiple optimization goals, and selects the optimal scheme through a multi-objective optimization algorithm. The optimization goals include but are not limited to cost, time, and quality;

The real-time execution and feedback adjustment module converts the optimal scheduling plan into specific execution instructions, and performs dynamic adjustments based on feedback by monitoring the execution process and results of the system in real time.

Furthermore, the data acquisition and preprocessing module includes the following components:

Sensor network: It consists of several wireless sensor nodes, which monitor and collect parameters or equipment status information of each link of service-oriented manufacturing through the nodes, and transmit the collected data to the data acquisition unit through wireless communication technology;

Data acquisition unit: receives the original sensory data and performs preliminary processing and organization. It may include some hardware devices, such as data acquisition cards, data converters, etc., as well as corresponding software programs to control the data acquisition process and ensure the accuracy and integrity of the data;

Data cleaning: Calculate the replacement values of missing values through statistical methods such as mean filling, median filling, and mode filling, remove or correct these bad data, and improve the quality and availability of data;

Data standardization: normalize the data through methods to convert the raw data into a uniform and comparable form;

Data storage: The cleaned and standardized data is stored for subsequent analysis and processing. The data storage involves the design and management of the database, as well as the formulation of data backup and recovery strategies.

Furthermore, the data standardization includes the following methods:

Minimum and maximum normalization: used to process different types of resource data. By compressing the original data into a specified range, the dimension differences between different data features are eliminated, so that the scheduling algorithm can accurately compare and evaluate the advantages and disadvantages of different resources, thereby making more reasonable scheduling decisions. The resource data includes but is not limited to equipment performance indicators, worker work efficiency, and order priority. The specific operations of the minimum and maximum normalization are as follows:

Map the original data to a specified range, usually [0,1] or [-1,1], using the conversion formula: [x_{\text{normalized}}＝\frac{x-x_{\text{min}}}{x_{\text{max}}-x_{\text{min}}}], where (x) is the original data, (x_{\text{min}}) and (x_{\text{max}}) are the minimum and maximum values in the data set, respectively;

Standard score method: used to process data with different distribution characteristics. By converting the original data into a standard normal distribution, the dimension and distribution differences between different data features are eliminated, making the scheduling algorithm stable and reliable when processing data. The specific operation of the standard score method is as follows:

The original data is converted into a standard normal distribution with a mean of 0 and a standard deviation of 1. The conversion formula is: [z＝\frac{x-\mu}{\sigma}], where (x) is the original data, (\mu) is the mean of the data, and (\sigma) is the standard deviation of the data.

Furthermore, the data analysis technology specifically includes:

Time series analysis: It is used to process data sequences arranged in chronological order. In a service-oriented manufacturing environment, it helps the system understand resource usage patterns, predict future demand trends, and identify abnormal events. It can predict future equipment maintenance needs through time series analysis of historical equipment failure data, thereby scheduling and arranging resources in advance.

Cluster analysis: used to group similar data points together. In manufacturing resource scheduling, it helps the system identify different types of resource usage patterns or workload patterns, and use these patterns to optimize resource allocation and scheduling to meet different production needs. Through cluster analysis of historical order data, order groups with similar characteristics can be identified, thereby formulating more accurate scheduling strategies.

Association rule mining: used to discover interesting relationships between data items. In service-oriented manufacturing, it helps the system identify potential associations between resource usage and their associations with production efficiency, cost and other goals, and use these association rules to guide resource scheduling and allocation to improve production efficiency and reduce costs. Association rule mining can be used to find that the use of certain equipment is highly correlated with specific types of orders, so that the allocation of these equipment can be given priority during scheduling.

Furthermore, the feature extraction includes: extracting key features using dimensionality reduction techniques, and calculating the correlation between the features and the target variable;

The dimensionality reduction technology specifically includes:

Principal component analysis: The original data is converted into a set of linearly unrelated variables, namely principal components, through orthogonal transformation to retain the main direction of change in the data while ignoring minor changes. In service-oriented manufacturing resource scheduling, it is used to reduce the dimension of features, remove redundant information, and retain the key change patterns in the data. The extracted principal components are used as the input of the subsequent learning algorithm to improve the performance and efficiency of the algorithm.

Autoencoder: used to learn data compression and encoding representation. It reconstructs input data by training a neural network and learns a low-dimensional hidden layer representation. In service-oriented manufacturing resource scheduling, it is used to extract feature representations of data. By training the autoencoder, the complex structure and pattern in the data are learned, and high-dimensional data is compressed into low-dimensional feature representations, which are then used as inputs for subsequent learning algorithms to optimize scheduling decisions.

The correlation between the calculated features and the target variable specifically includes:

Correlation analysis: used to quantify the linear relationship between features and target variables. By calculating the correlation coefficient between features and target variables, the system evaluates the degree of influence of features on target variables. In service-oriented manufacturing resource scheduling, it is used to identify features that are highly correlated with target variables. These features are of great significance for building prediction models and optimizing scheduling decisions.

Mutual information: It is used to measure the statistical correlation between two variables. It can capture not only linear relationships but also nonlinear relationships. In service-oriented manufacturing resource scheduling, it is used to evaluate the nonlinear association between features and target variables. By calculating the mutual information value between features and target variables, the system identifies features that are strongly correlated with the target variables, thereby guiding subsequent scheduling decisions.

Furthermore, the feature extraction steps are as follows:

Data exploration and understanding: Conduct in-depth exploration of raw data to understand its structure, distribution, and underlying patterns, including statistical description of data, visual presentation, and preliminary data cleaning and preprocessing.

Feature selection: Select features that have an impact on scheduling decisions based on business needs and problem background. The features include but are not limited to equipment performance indicators, historical fault data, worker work efficiency, and order characteristics.

Feature construction: When the original data does not contain features directly used for scheduling, new features need to be constructed based on business logic and domain knowledge;

Feature transformation: To improve the effectiveness and interpretability of features, the original features are transformed, including standardization, normalization, and encoding conversion;

Feature dimensionality reduction: When the number of features is too large, the dimensionality reduction technology is used to reduce the number of features to solve the problem of model overfitting and increased computational complexity, while retaining as much useful information as possible.

Furthermore, the adaptive learning algorithm training calculation content specifically includes:

Forward propagation: According to the parameters of the neural network, the value of the output layer is calculated through matrix multiplication and activation function;

Back propagation and optimization: Calculate the gradient according to the loss function and update the network parameters through optimization algorithms such as gradient descent, Adam, and RMSProp;

Hyperparameter tuning: Use grid search, random search, Bayesian optimization and other methods to tune hyperparameters such as learning rate, batch size, and number of iterations;

The steps of implementing the adaptive learning algorithm training module are as follows:

Data collection and processing: Collect historical and real-time data related to manufacturing resources, perform data cleaning, preprocessing and feature extraction to ensure data quality and availability;

Model selection and construction: According to the problem background and requirements, select appropriate deep learning, reinforcement learning and other algorithms to build a resource scheduling model;

Model training and optimization: Use historical data to train the model, calculate gradients through forward propagation and back propagation, update model parameters, and tune hyperparameters to improve model performance;

Model evaluation and verification: Use the verification set to evaluate the trained model and verify whether its performance meets the requirements. If not, further adjustment and optimization are required.

Real-time feedback and online learning: Deploy the trained model to the actual system, collect real-time feedback data, conduct online learning, and continuously update and optimize model parameters to adapt to changes in the environment;

Continuous monitoring and iteration: Continuously monitor the system to ensure the stability and performance of the model. At the same time, iteratively upgrade the model according to actual needs and technological development to maintain the advancement and competitiveness of the system.

Furthermore, the multi-objective optimization scheduling calculation content specifically includes:

Generate multiple scheduling solutions: Based on the output of the adaptive learning algorithm, multiple resource scheduling solutions are generated in combination with a heuristic algorithm, wherein the heuristic algorithm includes but is not limited to a genetic algorithm and a particle swarm optimization;

Scheme evaluation: Calculate the comprehensive score of each scheduling scheme by constructing a multi-objective evaluation function;

Select the optimal solution: Use a multi-objective optimization method to select the Pareto optimal solution as the final resource scheduling solution. The multi-objective optimization method includes but is not limited to the non-dominated sorting genetic algorithm NSGA-II and the multi-objective particle swarm optimization.

Furthermore, the real-time execution and feedback adjustment calculation content specifically includes:

Real-time scheduling calculation: quickly calculate and adjust resource scheduling plans based on real-time collected data and prediction results;

Feedback adjustment: Through online learning, incremental learning and other methods, the parameters of the adaptive learning model are updated in real time to adapt to environmental changes and new data.

Furthermore, the system includes a resource scheduling model, and the training and optimization of the resource scheduling model are based on an adaptive learning algorithm, specifically a hybrid model combining deep learning and reinforcement learning, and the training process of the model is expressed as the following formula:

State transition probability P(s_{t+1}|s_t,at): describes the probability of the next state s_{t+1} appearing given the current state s_t and the action a_t taken. In reinforcement learning, the state transition of the environment is usually determined by the dynamics of the environment, and the model needs to learn these dynamics to make better decisions;

Reward function R(s_t, a_t): represents the reward value obtained by the system given the current state (s_t) and action (a_t). The design of the reward function needs to take into account various factors in the service-oriented manufacturing process. This is the goal that the model tries to maximize during the learning process. The design of the reward function reflects the goal of the problem. The model will get a reward value based on the current state and action. This value will guide the model on how to make decisions in the subsequent learning;

Value function V(s_t): represents the expected total reward value that the model can obtain under a given state (s_t). This value is calculated by considering all possible paths starting from the current state and the corresponding rewards, and the value function is iteratively updated by the following formula:

(V(s_t)\leftarrow\max_{a_t}\left[R(s_t,a_t)+\gamma\sum_{s_{t+1}}P(s_{t+1}|s_t,a_t)V(s_{t+1})\right]), where \gamma is a discount factor used to balance immediate rewards and long-term rewards;

Policy function \pi(a_t|s_t): represents the probability distribution of selecting an action (a_t) given the current state (s_t). The update of the policy function is based on the difference between the reward and value functions, and the logarithmic gradient of the probability of selecting an action under the current policy. This process is actually adjusting the policy so that the probability of taking an action on a state-action pair with a higher reward increases, and the policy function is updated by the following formula:

(\pi(a_t|s_t)\leftarrow\pi(a_t|s_t)+\alpha\left[R(s_t,a_t)+\ga mma\sum_{s_{t+1}}P(s_{t+1}|s_t,a_t)V(s_{t+1})-V(s_t)\right]\frac{\partial\log\pi(a_t|s_t)}{\partial\pi(a_t|s_t)}), where \alpha is the learning rate, which is used to control the step size of the policy update.

The present invention provides a service-oriented manufacturing resource optimization scheduling system based on an adaptive learning algorithm, which has the following beneficial effects:

The present invention introduces an adaptive learning algorithm to enable the system to learn and optimize resource scheduling strategies in real time, thereby improving the efficiency of resource scheduling, and enhances the real-time adjustment capability of the system, so that the system can adjust the resource scheduling strategies in real time according to changes in the production environment to cope with various emergencies or changes in demand. At the same time, the prediction capability is improved. By analyzing and learning historical data, the system can predict future production needs and prepare for resource scheduling in advance. In addition, it makes full use of data resources, collects and analyzes data generated in the production process, extracts valuable information for optimizing resource scheduling strategies, and improves resource utilization and production efficiency.

Specifically, this digital service system has the following advantages:

1. Adaptability and flexibility: Since the system uses an adaptive learning algorithm, it can self-adjust and optimize according to the real-time manufacturing environment and resource status. Whether facing emergencies, equipment failures or changes in production demand, the system can respond quickly and adjust resource scheduling strategies to ensure efficient production. The excellent adaptability and flexibility enable the system to better adapt to the complex and changing manufacturing environment.

2. Improvement of resource utilization efficiency: By optimizing the scheduling algorithm, the system can accurately predict and allocate various manufacturing resources, such as equipment, manpower, materials, etc., which not only avoids waste and idleness of resources, but also ensures that the required resources can be obtained in time when needed, significantly improving resource utilization efficiency and reducing production costs.

3. Improved production efficiency and quality: As resources are more efficiently allocated and utilized, bottlenecks and delays in the production process are alleviated, which not only improves production efficiency and shortens production cycles, but also reduces errors and defects in the production process, thereby improving product quality.

4. Improvement of intelligence and automation levels: By introducing adaptive learning algorithms, intelligent and automated resource scheduling is achieved, which not only reduces dependence on manual intervention and the possibility of human errors, but also enables the system to continuously learn and improve, and constantly improve its own optimization capabilities.

5. Sustainability and environmental friendliness: By optimizing resource scheduling, the system helps reduce energy consumption and waste generation, and reduces the impact on the environment. At the same time, by improving production efficiency and product quality, it also helps to achieve sustainable manufacturing and production.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic diagram of the system architecture flow of a service-oriented manufacturing resource optimization and scheduling system based on an adaptive learning algorithm according to the present invention;

FIG2 is a schematic diagram of data collection and preprocessing of a service-oriented manufacturing resource optimization scheduling system based on an adaptive learning algorithm according to the present invention;

3 is a schematic diagram of data analysis and feature extraction of a service-oriented manufacturing resource optimization scheduling system based on an adaptive learning algorithm according to the present invention;

4 is a schematic diagram of an adaptive learning algorithm training module of a service-oriented manufacturing resource optimization scheduling system based on an adaptive learning algorithm according to the present invention;

5 is a schematic diagram of a multi-objective optimization scheduling module of a service-oriented manufacturing resource optimization scheduling system based on an adaptive learning algorithm according to the present invention;

FIG6 is a schematic diagram of a real-time execution and feedback adjustment module of a service-oriented manufacturing resource optimization scheduling system based on an adaptive learning algorithm according to the present invention.

DETAILED DESCRIPTION

The following embodiments of the present invention are described in further detail in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

As shown in Figures 1 to 6, the service-oriented manufacturing resource optimization scheduling system based on the adaptive learning algorithm includes:

The data acquisition and preprocessing module is used to collect raw data and perform preprocessing operations on the data; the data acquisition and preprocessing module includes the following components:

Sensor network: It consists of several wireless sensor nodes, which monitor and collect parameters or equipment status information of each link of service-oriented manufacturing through the nodes, and transmit the collected data to the data acquisition unit through wireless communication technology;

Data acquisition unit: receives the original sensory data and performs preliminary processing and organization. It may include some hardware devices, such as data acquisition cards, data converters, etc., as well as corresponding software programs to control the data acquisition process and ensure the accuracy and integrity of the data;

Data cleaning: Calculate the replacement values of missing values through statistical methods such as mean filling, median filling, and mode filling, remove or correct these bad data, and improve the quality and availability of data;

Data standardization: normalize the data through methods to convert the raw data into a uniform and comparable form;

Data storage: Storing cleaned and standardized data for subsequent analysis and processing. The data storage involves database design and management, as well as the formulation of data backup and recovery strategies.

The data standardization includes the following methods:

Minimum and maximum normalization: used to process different types of resource data. By compressing the original data into a specified range, the dimension differences between different data features are eliminated, so that the scheduling algorithm can accurately compare and evaluate the advantages and disadvantages of different resources, thereby making more reasonable scheduling decisions. The resource data includes but is not limited to equipment performance indicators, worker work efficiency, and order priority. The specific operations of the minimum and maximum normalization are as follows:

Map the original data to a specified range, usually [0,1] or [-1,1], using the conversion formula: [x_{\text{normalized}}＝\frac{x-x_{\text{min}}}{x_{\text{max}}-x_{\text{min}}}], where (x) is the original data, (x_{\text{min}}) and (x_{\text{max}}) are the minimum and maximum values in the data set, respectively;

Standard score method: used to process data with different distribution characteristics. By converting the original data into a standard normal distribution, the dimension and distribution differences between different data features are eliminated, making the scheduling algorithm stable and reliable when processing data. The specific operation of the standard score method is as follows:

The original data is converted into a standard normal distribution with a mean of 0 and a standard deviation of 1. The conversion formula is: [z＝\frac{x-\mu}{\sigma}], where (x) is the original data, (\mu) is the mean of the data, and (\sigma) is the standard deviation of the data;

A data analysis and feature extraction module, which uses data analysis technology to extract key features from the data for training the adaptive learning model;

The data analysis technology specifically includes:

Time series analysis: It is used to process data sequences arranged in chronological order. In a service-oriented manufacturing environment, it helps the system understand resource usage patterns, predict future demand trends, and identify abnormal events. It can predict future equipment maintenance needs through time series analysis of historical equipment failure data, thereby scheduling and arranging resources in advance.

Cluster analysis: used to group similar data points together. In manufacturing resource scheduling, it helps the system identify different types of resource usage patterns or workload patterns, and use these patterns to optimize resource allocation and scheduling to meet different production needs. Through cluster analysis of historical order data, order groups with similar characteristics can be identified, thereby formulating more accurate scheduling strategies.

Association rule mining: used to discover interesting relationships between data items. In service-oriented manufacturing, it helps the system identify potential associations between resource usage and their associations with production efficiency, cost and other goals, and use these association rules to guide resource scheduling and allocation to improve production efficiency and reduce costs. Through association rule mining, it can be found that the use of certain equipment is highly correlated with specific types of orders, so that the allocation of these equipment can be given priority during scheduling;

The feature extraction includes: extracting key features using dimensionality reduction technology, and calculating the correlation between the features and the target variable;

The dimensionality reduction technology specifically includes:

Principal component analysis: The original data is converted into a set of linearly unrelated variables, namely principal components, through orthogonal transformation to retain the main direction of change in the data while ignoring minor changes. In service-oriented manufacturing resource scheduling, it is used to reduce the dimension of features, remove redundant information, and retain the key change patterns in the data. The extracted principal components are used as the input of the subsequent learning algorithm to improve the performance and efficiency of the algorithm.

Autoencoder: used to learn data compression and encoding representation. It reconstructs input data by training a neural network and learns a low-dimensional hidden layer representation. In service-oriented manufacturing resource scheduling, it is used to extract feature representations of data. By training the autoencoder, the complex structure and pattern in the data are learned, and high-dimensional data is compressed into low-dimensional feature representations, which are then used as inputs for subsequent learning algorithms to optimize scheduling decisions.

The correlation between the calculated features and the target variable specifically includes:

Correlation analysis: used to quantify the linear relationship between features and target variables. By calculating the correlation coefficient between features and target variables, the system evaluates the degree of influence of features on target variables. In service-oriented manufacturing resource scheduling, it is used to identify features that are highly correlated with target variables. These features are of great significance for building prediction models and optimizing scheduling decisions.

Mutual information: It is used to measure the statistical correlation between two variables. It can capture not only linear relationships but also nonlinear relationships. In service-oriented manufacturing resource scheduling, it is used to evaluate the nonlinear association between features and target variables. By calculating the mutual information value between features and target variables, the system identifies features that are strongly correlated with the target variable, thereby guiding subsequent scheduling decisions.

The feature extraction steps are as follows:

Data exploration and understanding: Conduct in-depth exploration of raw data to understand its structure, distribution, and underlying patterns, including statistical description of data, visual presentation, and preliminary data cleaning and preprocessing.

Feature selection: Select features that have an impact on scheduling decisions based on business needs and problem background. The features include but are not limited to equipment performance indicators, historical fault data, worker work efficiency, and order characteristics.

Feature construction: When the original data does not contain features directly used for scheduling, new features need to be constructed based on business logic and domain knowledge;

Feature transformation: To improve the effectiveness and interpretability of features, the original features are transformed, including standardization, normalization, and encoding conversion;

Feature dimensionality reduction: When there are too many features, the dimensionality reduction technology is used to reduce the number of features to solve the problem of model overfitting and increased computational complexity, while retaining as much useful information as possible;

Adaptive learning algorithm training module, which uses deep learning and reinforcement learning algorithms, combined with historical data and real-time feedback, to continuously train and optimize resource scheduling models; the adaptive learning algorithm training calculation content specifically includes:

Forward propagation: According to the parameters of the neural network, the value of the output layer is calculated through matrix multiplication and activation function;

Back propagation and optimization: Calculate the gradient according to the loss function and update the network parameters through optimization algorithms such as gradient descent, Adam, and RMSProp;

Hyperparameter tuning: Use grid search, random search, Bayesian optimization and other methods to tune hyperparameters such as learning rate, batch size, and number of iterations;

The steps of implementing the adaptive learning algorithm training module are as follows:

Data collection and processing: Collect historical and real-time data related to manufacturing resources, perform data cleaning, preprocessing and feature extraction to ensure data quality and availability;

Model selection and construction: According to the problem background and requirements, select appropriate deep learning, reinforcement learning and other algorithms to build a resource scheduling model;

Model training and optimization: Use historical data to train the model, calculate gradients through forward propagation and back propagation, update model parameters, and tune hyperparameters to improve model performance;

Model evaluation and verification: Use the verification set to evaluate the trained model and verify whether its performance meets the requirements. If not, further adjustment and optimization are required.

Real-time feedback and online learning: Deploy the trained model to the actual system, collect real-time feedback data, conduct online learning, and continuously update and optimize model parameters to adapt to changes in the environment;

Continuous monitoring and iteration: Continuously monitor the system to ensure the stability and performance of the model. At the same time, iterate and upgrade the model according to actual needs and technological development to maintain the advancement and competitiveness of the system;

A multi-objective optimization scheduling module, based on the output of the adaptive learning algorithm, considers multiple optimization objectives, generates multiple resource scheduling solutions, and selects the optimal solution through a multi-objective optimization algorithm. The optimization objectives include but are not limited to cost, time, and quality. The multi-objective optimization scheduling calculation content specifically includes:

Generate multiple scheduling solutions: Based on the output of the adaptive learning algorithm, multiple resource scheduling solutions are generated in combination with a heuristic algorithm, wherein the heuristic algorithm includes but is not limited to a genetic algorithm and a particle swarm optimization;

Scheme evaluation: Calculate the comprehensive score of each scheduling scheme by constructing a multi-objective evaluation function;

Select the optimal solution: Use a multi-objective optimization method to select the Pareto optimal solution as the final resource scheduling solution. The multi-objective optimization method includes but is not limited to the non-dominated sorting genetic algorithm NSGA-II and the multi-objective particle swarm optimization;

A real-time execution and feedback adjustment module converts the optimal scheduling plan into specific execution instructions, and dynamically adjusts according to the feedback by real-time monitoring of the system execution process and results; the real-time execution and feedback adjustment calculation content specifically includes:

Real-time scheduling calculation: quickly calculate and adjust resource scheduling plans based on real-time collected data and prediction results;

Feedback adjustment: Through online learning, incremental learning and other methods, the parameters of the adaptive learning model are updated in real time to adapt to changes in the environment and new data;

The system includes a resource scheduling model, and the training and optimization of the resource scheduling model are based on an adaptive learning algorithm, specifically a hybrid model combining deep learning and reinforcement learning, and the training process of the model is expressed as the following formula:

State transition probability P(s_{t+1}|s_t,at): describes the probability of the next state s_{t+1} appearing given the current state s_t and the action a_t taken. In reinforcement learning, the state transition of the environment is usually determined by the dynamics of the environment, and the model needs to learn these dynamics to make better decisions;

Reward function R(s_t, a_t): represents the reward value obtained by the system given the current state (s_t) and action (a_t). The design of the reward function needs to take into account various factors in the service-oriented manufacturing process. This is the goal that the model tries to maximize during the learning process. The design of the reward function reflects the goal of the problem. The model will get a reward value based on the current state and action. This value will guide the model on how to make decisions in the subsequent learning;

Value function V(s_t): represents the expected total reward value that the model can obtain under a given state (s_t). This value is calculated by considering all possible paths starting from the current state and the corresponding rewards, and the value function is iteratively updated by the following formula:

(V(s_t)\leftarrow\max_{a_t}\left[R(s_t,a_t)+\gamma\sum_{s_{t+1}}P(s_{t+1}|s_t,a_t)V(s_{t+1})\right]), where \gamma is a discount factor used to balance immediate rewards and long-term rewards;

Policy function \pi(a_t|s_t): represents the probability distribution of selecting an action (a_t) given the current state (s_t). The update of the policy function is based on the difference between the reward and value functions and the logarithmic gradient of the probability of selecting an action under the current policy. This process is actually adjusting the policy so that the probability of taking an action on a state-action pair with a higher reward increases, and the policy function is updated by the following formula:

(\pi(a_t|s_t)\leftarrow\pi(a_t|s_t)+\alpha\left[R(s_t,a_t)+\ga mma\sum_{s_{t+1}}P(s_{t+1}|s_t,a_t)V(s_{t+1})-V(s_t)\right]\frac{\partial\log\pi(a_t|s_t)}{\partial\pi(a_t|s_t)}), where \alpha is the learning rate, which is used to control the step size of the policy update;

The following is a simplified code example that shows how to combine the updates of the value function and the policy function to train the resource scheduling model:

# Initialize the value function V and the strategy function pi

V＝initialize_value_function()

pi = initialize_policy()

#Training loop

for episode in range(num_episodes):

#Reset environment status s_t

s_t = env.reset()

#Iterate in an episode

while not env.is_done(s_t):

#Select action a_t according to the current strategy

a_t＝pi.choose_action(s_t)

#Execute action a_t, observe reward r_t and next state s_{t+1}

r_t,s_{t+1}＝env.step(a_t)

# Update value function V

V.update(s_t,r_t,s_{t+1})

#Update strategy function pi

pi.update(s_t,a_t,r_t,s_{t+1},V)

#Transfer to the next state s_{t+1}

s_t＝s_{t+1}

#After training is completed, use the final strategy function for resource scheduling.

The embodiments of the present invention are given for the purpose of illustration and description, and are not intended to be exhaustive or to limit the invention to the disclosed forms. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present invention and to enable those of ordinary skill in the art to understand the present invention and thereby design various embodiments with various modifications suitable for specific uses.

Claims (10)

1. The service type manufacturing resource optimization scheduling system based on the self-adaptive learning algorithm is characterized by comprising the following components:

The data acquisition and preprocessing module is used for collecting original data and preprocessing the data;

The data analysis and feature extraction module is used for extracting key features in the data by utilizing a data analysis technology and is used for training the self-adaptive learning model;

The self-adaptive learning algorithm training module adopts a deep learning and reinforcement learning algorithm, combines historical data and real-time feedback, and continuously trains and optimizes a resource scheduling model;

The multi-objective optimization scheduling module is used for generating a plurality of sets of resource scheduling schemes by considering a plurality of optimization objectives based on the output of the self-adaptive learning algorithm and selecting an optimal scheme through the multi-objective optimization algorithm, wherein the optimization objectives comprise but are not limited to cost, time and quality;

and the real-time execution and feedback adjustment module converts the optimal scheduling scheme into a specific execution instruction, and dynamically adjusts according to feedback by monitoring the execution process and the result of the system in real time.

2. The adaptive learning algorithm-based service-oriented manufacturing resource optimization scheduling system of claim 1, wherein the data acquisition and preprocessing module comprises:

Sensor network: the system comprises a plurality of wireless sensor nodes, wherein parameters or equipment state information of each link of service type manufacturing are monitored and collected through the nodes, and collected data are transmitted to a data collection unit through a wireless communication technology;

a data acquisition unit: the method comprises the steps of receiving original perception data, and primarily processing and arranging the original perception data, wherein the original perception data possibly comprises hardware devices such as a data acquisition card, a data converter and the like, and corresponding software programs for controlling the data acquisition process so as to ensure the accuracy and the completeness of the data;

Data cleaning: calculating substitution values of the missing values through statistical methods such as mean filling, median filling and mode filling, removing or correcting the bad data, and improving the quality and usability of the data;

Data normalization: normalizing the data by a method to convert the original data into a unified and comparable form;

and (3) data storage: the cleaned and standardized data is stored for later analysis and processing, which involves the design and management of databases and the formulation of data backup and restore policies.

3. The adaptive learning algorithm-based service-oriented manufacturing resource optimization scheduling system of claim 2, wherein the data normalization comprises the following method:

Minimum maximum normalization: the method is used for processing different types of resource data, the dimension difference between different data characteristics is eliminated by compressing the original data to a specified range, so that a scheduling algorithm can accurately compare and evaluate the advantages and disadvantages of different resources, thereby making more reasonable scheduling decisions, the resource data comprises but is not limited to equipment performance indexes, worker work efficiency and order priority, and the specific operations of the minimum maximum normalization are as follows:

Mapping the raw data to a specified range, typically [0,1] or [ -1,1], conversion formula: [ x_ { _text { normalized } = \frac { x_ { \text { min } { x_ { \text { max } } } } } } } } -wherein, (x) is the raw data, (x_ { _text { min }) and (x_text { max }) are the minimum and maximum values, respectively, in the dataset;

Standard score method: the method is used for processing the data with different distribution characteristics, and the dimension and distribution difference between different data characteristics is eliminated by converting the original data into a standard normal distribution form, so that a scheduling algorithm is stable and reliable in processing the data, and the specific operation of the standard score method is as follows:

Converting the original data into standard normal distribution with the mean value of 0 and the standard deviation of 1, and converting the formula: [ z= \frac { x\mu } { \sigma } ], where (x) is raw data, (\mu) is the mean value of the data and (\sigma) is the standard deviation of the data.

4. The adaptive learning algorithm-based service-oriented manufacturing resource optimization scheduling system of claim 1, wherein the data analysis technique specifically comprises:

Time series analysis: the method is used for processing the data sequences arranged in time sequence, helping a system to understand the use mode of resources, predicting future demand trend and identifying abnormal events in a service type manufacturing environment, and the future maintenance requirement of the equipment can be predicted through time sequence analysis of historical equipment fault data so as to schedule and arrange the resources in advance;

and (3) cluster analysis: the method is used for grouping similar data points together, in the process of scheduling manufacturing resources, helping a system identify different types of resource use modes or workload modes, optimizing allocation and scheduling of the resources by utilizing the modes so as to meet different production demands, and identifying order groups with similar characteristics through cluster analysis of historical order data so as to formulate a more accurate scheduling strategy;

Association rule mining: for discovering interesting relationships between data items, in service-type manufacturing, a help system identifies potential associations between resource uses and their associations with targets of production efficiency, cost, etc., and uses these association rules to guide scheduling and configuration of resources to improve production efficiency, reduce cost, and can discover that the use of certain devices is highly correlated with orders of a specific type through association rule mining, thereby giving priority to allocation of those devices in scheduling.

5. The adaptive learning algorithm-based service-oriented manufacturing resource optimization scheduling system of claim 1, wherein the feature extraction comprises: extracting key features by using a dimension reduction technology, and calculating the relevance between the features and the target variable;

the dimension reduction technology specifically comprises the following steps:

And (3) principal component analysis: the original data is converted into a group of linear uncorrelated variables, namely main components, so as to preserve the main change direction in the data and ignore the secondary change, and the method is used for reducing the feature dimension, removing redundant information and preserving the key change mode in the data in service type manufacturing resource scheduling, and the extracted main components are used as the input of a subsequent learning algorithm to improve the performance and efficiency of the algorithm;

self-encoder: the method comprises the steps of training a neural network to reconstruct input data, simultaneously learning a low-dimensional hidden layer representation, extracting characteristic representation of the data in service type manufacturing resource scheduling, learning complex structures and modes in the data by training a self-encoder, compressing high-dimensional data into the low-dimensional characteristic representation, and optimizing scheduling decisions as input of a subsequent learning algorithm;

The correlation between the calculation feature and the target variable specifically comprises:

Correlation analysis: the method is used for quantifying the linear relation between the features and the target variables, evaluating the influence degree of the features on the target variables by the system through calculating the correlation coefficient between the features and the target variables, and identifying the features highly correlated with the target variables in service type manufacturing resource scheduling, wherein the features have important significance for constructing a prediction model and optimizing scheduling decisions;

Mutual information: the method is used for measuring the statistical correlation between two variables, can capture not only a linear relationship but also a nonlinear relationship, is used for evaluating the nonlinear correlation between the characteristics and the target variables in service type manufacturing resource scheduling, and is used for identifying the characteristics with strong correlation with the target variables by calculating the mutual information values between the characteristics and the target variables so as to guide the subsequent scheduling decision.

6. The adaptive learning algorithm-based service-oriented manufacturing resource optimization scheduling system of claim 5, wherein the feature extraction step is as follows:

Data exploration and understanding: deep exploration is carried out on the original data, and the structure, distribution and potential rules of the original data are understood, including statistical description, visual presentation and preliminary data cleaning and preprocessing of the data;

Feature selection: selecting characteristics which have an influence on scheduling decisions according to business requirements and problem backgrounds, wherein the characteristics comprise, but are not limited to, performance indexes of equipment, historical fault data, working efficiency of workers and characteristics of orders;

characteristic structure: when the original data does not contain the features directly used for scheduling, new features need to be constructed according to business logic and domain knowledge;

Feature transformation: to improve the effectiveness and the interpretability of the features, the original features are transformed, including standardization, normalization and transcoding;

Feature dimension reduction: when the number of the features is excessive, the number of the features is reduced through a dimension reduction technology for the problems of model overfitting and calculation complexity increase, and meanwhile, as much useful information as possible is reserved.

7. The adaptive learning algorithm-based service-type manufacturing resource optimization scheduling system according to claim 1, wherein the adaptive learning algorithm training calculation content specifically comprises:

Forward propagation: calculating the value of an output layer through matrix multiplication and activation functions according to the parameters of the neural network;

Back propagation and optimization: calculating a gradient according to the loss function, and updating network parameters through optimization algorithms such as gradient descent, adam, RMSProp and the like;

Super parameter tuning: the super parameters such as the learning rate, the batch size, the iteration number and the like are optimized by using the methods such as grid search, random search and Bayesian optimization;

the training module for implementing the self-adaptive learning algorithm comprises the following steps:

Data collection and processing: collecting historical data and real-time data related to manufacturing resources, and performing data cleaning, preprocessing and feature extraction to ensure the quality and usability of the data;

model selection and construction: selecting proper algorithms such as deep learning, reinforcement learning and the like according to the problem background and the requirements, and constructing a resource scheduling model;

Model training and optimizing: model training is carried out by using historical data, gradient is calculated through forward propagation and backward propagation, model parameters are updated, and super parameters are adjusted to improve the performance of the model;

Model evaluation and verification: evaluating the trained model by using a verification set, verifying whether the performance of the model meets the requirements, and if not, further adjusting and optimizing the model;

real-time feedback and online learning: deploying the trained model into an actual system, collecting real-time feedback data, and performing online learning to continuously update and optimize model parameters so as to adapt to the change of the environment;

Continuously monitoring and iterating: the system is continuously monitored, the stability and the performance of the model are ensured, and meanwhile, the model is iteratively upgraded according to actual requirements and technical development, so that the advancement and the competitiveness of the system are maintained.

8. The adaptive learning algorithm-based service-type manufacturing resource optimization scheduling system according to claim 1, wherein the multi-objective optimization scheduling calculation content specifically includes:

generating a plurality of sets of scheduling schemes: based on the output of the adaptive learning algorithm, generating a plurality of resource scheduling schemes in combination with a heuristic algorithm, wherein the heuristic algorithm comprises, but is not limited to, a genetic algorithm and particle swarm optimization;

Scheme evaluation: calculating the comprehensive score of each scheduling scheme by constructing a multi-objective evaluation function;

Selecting an optimal scheme: the Pareto optimal solution is selected as the final resource scheduling scheme using a multi-objective optimization method including, but not limited to, non-dominant ranking genetic algorithm NSGA-II, multi-objective particle swarm optimization.

9. The adaptive learning algorithm-based service-oriented manufacturing resource optimization scheduling system of claim 1, wherein the real-time execution and feedback adjustment calculation content specifically comprises:

and (3) real-time scheduling calculation: according to the data collected in real time and the prediction result, a resource scheduling scheme is rapidly calculated and adjusted;

and (3) feedback adjustment: parameters of the self-adaptive learning model are updated in real time by methods of online learning, incremental learning and the like so as to adapt to environmental changes and new data.

10. The adaptive learning algorithm-based service-oriented manufacturing resource optimization scheduling system according to claim 1, wherein the system comprises a resource scheduling model, and the training and optimization of the resource scheduling model is based on the adaptive learning algorithm, in particular a hybrid model combining deep learning and reinforcement learning, and the training process of the model is expressed as the following formula:

state transition probability P (s_ { t+1} |s_t, a_t): describing the likelihood that the next state s_ { t+1} will occur given the current state s_t and the action taken a_t, in reinforcement learning, the state transitions of the environment are typically determined by the dynamics of the environment, which the model needs to learn to make better decisions;

Bonus function R (s_t, a_t): representing a prize value obtained by the system given the current state (s_t) and the action (a_t), the design of the prize function taking into account various factors in the service-type manufacturing process, which is the goal of the model in trying to maximize during learning, the design of the prize function reflecting the goal of the problem, the model deriving a prize value from the current state and action, which value directs the model in how to make decisions in the subsequent learning;

Value function V (s_t): representing the expected total prize value that the model can obtain in a given state (s_t), this value being calculated by taking into account all possible paths from the current state and the corresponding prizes, and the value function being iteratively updated by the following formula:

(V(s_t)\leftarrow\max_{a_t}\left[R(s_t,a_t)+\gamma\sum_{s_{t+1}}P(s_{t+1}|s_t,a_t)V(s_{t+1})\right]), Wherein gamma is a discount factor for balancing instant rewards and long-term rewards;

Policy function \pi (a_t|s_t): representing a probability distribution of a selected action (a_t) given a current state (s_t), the updating of the policy function is based on the differences of the rewards and the value functions, and the logarithmic gradient of the probability of the selected action under the current policy, this process is actually adjusting the policy such that the probability of taking an action on the state-action pair that gets a higher rewards increases, and the policy function is updated by the following formula:

(\pi(a_t|s_t)\leftarrow\pi(a_t|s_t)+\alpha\left[R(s_t,a_t)+\ga mma\sum_{s_{t+1}}P(s_{t+1}|s_t,a_t)V(s_{t+1})-V(s_t)\right]\frac{\partial\log\pi(a_t|s_t)}{\partial\pi(a_t|s_t)}), Where \alpha is the learning rate, used to control the step size of the policy update.