CN118822229B

CN118822229B - Power grid energy storage dispatching method and system based on deep learning and energy management

Info

Publication number: CN118822229B
Application number: CN202411308809.0A
Authority: CN
Inventors: 徐诚; 李双; 向阳; 邓琪; 李建邦
Original assignee: Wuhan Yichen Chuangxiang Technology Co ltd
Current assignee: Wuhan Yichen Chuangxiang Technology Co ltd
Priority date: 2024-09-19
Filing date: 2024-09-19
Publication date: 2024-11-29
Anticipated expiration: 2044-09-19
Also published as: CN118822229A

Abstract

The invention provides a power grid energy storage scheduling method and system based on deep learning and energy management, wherein the method comprises the following steps of obtaining historical electricity utilization data and real-time electricity utilization data of an electricity utilization load; the method comprises the steps of constructing an electricity utilization characteristic identification model based on a deep convolutional neural network, analyzing real-time electricity utilization data by using the electricity utilization characteristic identification model, predicting to obtain the predicted electricity utilization characteristics of an electricity utilization load, calculating to obtain the predicted electricity utilization total amount of a power distribution area, constructing a power generation prediction model based on a support vector machine, preprocessing weather prediction data, inputting the preprocessed weather prediction data into the power generation prediction model to obtain predicted surplus electric energy of a micro-grid, calculating to obtain predicted energy storage data of a shared energy storage device, constructing an energy scheduling block chain network of the power distribution network based on an energy storage server, combining all the predicted energy storage data, and generating an electric energy scheduling scheme of the power distribution network through the energy scheduling block chain network. The invention has the effect of reasonably scheduling the energy storage of the power grid.

Description

Power grid energy storage scheduling method and system based on deep learning and energy management

Technical Field

The invention belongs to the technical field of power grid energy storage scheduling, and particularly relates to a power grid energy storage scheduling method and system based on deep learning and energy management.

Background

Conventional grid systems typically are based on centralized power generation, with power being transmitted from a power plant to each power terminal, so that power scheduling for the conventional grid system only requires a simple planned scheduling scheme. With the development and application of renewable energy technologies, particularly the rapid growth of distributed energy sources such as photovoltaic power generation and wind power generation, the traditional power grid system faces great reform and challenges.

Photovoltaic power generation and wind power generation have extremely strong instability and volatility, and the distribution range is wide. In order to avoid the waste of energy generated by photovoltaic power generation and wind power generation, energy storage devices are often configured for the photovoltaic power generation area and the wind power generation area to store the electric energy overflowed from the corresponding areas. Because the traditional planning and scheduling scheme only depends on the power generation data of the power plant and a small amount of terminal power consumption data to generate, the traditional scheduling scheme has difficulty in coping with complicated and changeable power demands and supply conditions, so that the operation efficiency of a power grid is low, and the energy waste is serious. Therefore, the concept of smart grids has been developed, and smart grids aim to realize efficient, reliable and intelligent operation of the grids through information communication technology and advanced control means. In this context, how to effectively manage and schedule distributed energy sources and energy storage devices is one of the core problems of smart grid research and applications.

Disclosure of Invention

The invention provides a power grid energy storage scheduling method and system based on deep learning and energy management, which are used for solving the problems of low operation efficiency and serious energy waste of a power grid containing a large amount of distributed new energy sources caused by a traditional power grid scheduling mode.

In a first aspect, the present invention provides a power grid energy storage scheduling method based on deep learning and energy management, applied to a power distribution network, where the power distribution network includes a plurality of power distribution areas, each power distribution area includes a plurality of power loads and a plurality of micro-grids constructed based on a photovoltaic wind power generation system, each power distribution area is further configured with a shared energy storage device, where the shared energy storage device is used to store surplus energy produced by all the micro-grids in the power distribution area, and the surplus energy represents a portion of the energy produced by the micro-grids in the same time period that exceeds the energy supplied by the micro-grids, and the method includes the following steps:

for each power distribution area, acquiring historical power utilization data and real-time power utilization data of all the power utilization loads in the power distribution area;

constructing an electricity utilization characteristic recognition model based on a deep convolutional neural network and through the historical electricity utilization data, analyzing the real-time electricity utilization data by utilizing the electricity utilization characteristic recognition model, and predicting to obtain the predicted electricity utilization characteristic of the electricity utilization load;

Calculating to obtain the predicted electricity consumption total amount of the power distribution area according to the predicted electricity consumption characteristics of all the electricity consumption loads;

Acquiring historical weather data and weather forecast data of the power distribution area and historical surplus electric energy data of all the micro-grids, wherein the historical weather data and the historical surplus electric energy data are in the same historical time period;

based on a support vector machine, constructing a power generation prediction model through the historical weather data and the historical surplus power data, preprocessing the weather prediction data, and inputting the preprocessed weather prediction data into the power generation prediction model to obtain the predicted surplus power of the micro-grid;

Calculating to obtain predicted energy storage data of the shared energy storage device by combining the predicted total power consumption, the predicted surplus electric energy of all the micro-grids and the residual stored electric energy of the shared energy storage device, and storing the predicted energy storage data in an energy storage server carried by the shared energy storage device;

And constructing an energy scheduling blockchain network of the power distribution network based on the energy storage server of each shared energy storage device, combining all the predicted energy storage data and generating an electric energy scheduling scheme of the power distribution network through the energy scheduling blockchain network.

Optionally, the method for constructing the electricity feature recognition model based on the deep convolutional neural network and through the historical electricity data, analyzing the real-time electricity data by using the electricity feature recognition model, and predicting to obtain the predicted electricity feature of the electricity load includes the following steps:

Constructing an initial electricity utilization characteristic recognition model based on a deep convolutional neural network;

Preprocessing all the historical electricity utilization data into a historical electricity utilization data set, wherein all the historical electricity utilization data are marked with corresponding historical electricity utilization characteristics in advance;

performing feature correlation analysis on the historical electricity utilization features of the historical electricity utilization data set to obtain an electricity utilization feature classification matrix of the historical electricity utilization data set;

Training the initial electricity utilization characteristic recognition model by using the electricity utilization characteristic classification matrix to obtain a trained electricity utilization characteristic recognition model;

And analyzing the real-time electricity utilization data by using the electricity utilization characteristic identification model, and predicting to obtain the predicted electricity utilization characteristics of the electricity utilization load.

Optionally, the performing feature correlation analysis on the historical electricity utilization features of the historical electricity utilization data set, and obtaining the electricity utilization feature classification matrix of the historical electricity utilization data set includes the following steps:

Optionally, extracting the historical electricity utilization feature of the historical electricity utilization dataset;

Calculating by adopting a connection function to obtain the electricity utilization feature association degree of the historical electricity utilization features of different feature types;

And analyzing the power utilization characteristic association degree by utilizing a decision tree learning algorithm based on information entropy, and acquiring a power utilization characteristic classification matrix of the historical power utilization data set.

Optionally, the training the initial electricity feature recognition model by using the electricity feature classification matrix, and obtaining the trained electricity feature recognition model includes the following steps:

selecting optimal electricity utilization characteristics from the electricity utilization characteristic classification matrix by adopting a particle swarm optimization algorithm;

constructing an optimal feature matrix based on the optimal electricity utilization feature through a weighted least square method;

Inputting the optimal feature matrix into the initial electricity utilization feature recognition model, and calculating to obtain bias item parameters of all the full-connection layers in the initial electricity utilization feature recognition model;

calculating a parameter error of the bias term parameter through a preset optimal parameter;

If the parameter error is greater than or equal to a preset error threshold, all the steps are re-executed until the parameter error is smaller than the error threshold;

And if the parameter error is smaller than the error threshold, completing the model training process of the initial electricity utilization characteristic recognition model to obtain a trained electricity utilization characteristic recognition model.

Optionally, the selecting the optimal electricity utilization feature from the electricity utilization feature classification matrix by adopting a particle swarm optimization algorithm includes the following steps:

Randomly selecting initial electricity utilization characteristics from the electricity utilization characteristic classification matrix;

Initializing the initial electricity utilization characteristic into two-dimensional characteristic data by using preset particle swarm algorithm parameters, and constructing an initial particle swarm according to the two-dimensional characteristic data;

performing repeated iterative optimization on the initial particle swarm data through a particle swarm optimization algorithm until the initial particle swarm is optimized to be an optimal particle swarm;

And taking all the initial electricity utilization characteristics in the optimal particle swarm as optimal electricity utilization characteristics.

Optionally, the step of constructing a power generation prediction model based on the support vector machine and through the historical weather data and the historical surplus electrical energy data, preprocessing the weather prediction data, and inputting the preprocessed weather prediction data into the power generation prediction model to obtain the predicted surplus electrical energy of the micro-grid includes the following steps:

Constructing an initial power generation prediction model based on a support vector machine;

respectively calculating the data association degree between the historical weather data and the historical surplus electric energy data of different types by using a gray association algorithm;

integrating all the historical weather data with the data association degree larger than a preset association degree threshold value into a historical weather data set;

performing iterative training on the initial power generation prediction model by combining the historical weather data set and the historical surplus power data until model parameters in the initial power generation prediction model are trained to optimal model parameters, so as to obtain a power generation prediction model after training;

And preprocessing the weather prediction data, and inputting the preprocessed weather prediction data into the power generation prediction model to obtain the predicted surplus electric energy of the micro-grid.

Optionally, the preprocessing the weather prediction data, inputting the preprocessed weather prediction data into the power generation prediction model, and obtaining the predicted surplus electric energy of the micro-grid includes the following steps:

extracting all weather data types in the historical weather data set;

screening out invalid weather prediction data with different data types from all the weather prediction data types;

And inputting all the remaining weather prediction data into the power generation prediction model to obtain the predicted surplus electric energy of the micro-grid.

Optionally, the energy scheduling block chain network of the power distribution network is built based on the energy storage server of each shared energy storage device, and the power scheduling scheme for combining all the predicted energy storage data and generating the power distribution network through the energy scheduling block chain network includes the following steps:

Constructing an energy scheduling blockchain network of the power distribution network based on the energy storage server of each shared energy storage device;

according to the server states of the energy storage servers, electing a leading server in all the energy storage servers;

marking the shared energy storage device of which the predicted energy storage data is lower than a preset electric energy data threshold value as a target shared energy storage device, and marking the energy storage server of the target shared energy storage device as a target energy storage server;

Automatically generating an electric energy scheduling request of the target shared energy storage device through the target energy storage server, and uploading the electric energy scheduling request to the energy scheduling block chain network;

Broadcasting the power scheduling request to other energy storage servers except the target energy storage server by utilizing the leading server;

When the energy storage server receives the electric energy scheduling request, generating electric energy scheduling information by combining the electric energy scheduling request and the locally stored predicted energy storage data, and uploading the electric energy scheduling information to the energy scheduling block chain network;

and summarizing all the electric energy scheduling information through the leading server and generating an electric energy scheduling scheme of the power distribution network.

Optionally, the step of summarizing all the power scheduling information by the leader server and generating a power scheduling scheme of the power distribution network includes the following steps:

summarizing all the electric energy scheduling information through the leading server and generating an initial electric energy scheduling scheme of the power distribution network;

broadcasting the initial power scheduling scheme to all the energy storage servers through the lead server;

Verifying whether the initial power scheduling scheme passes verification by using a consensus mechanism;

If the verification of the initial power dispatching scheme is not passed, acquiring all the power dispatching information regenerated by the energy storage server through the energy dispatching block chain network, and repeatedly executing the scheme generation step and the scheme verification step until the verification of the initial power dispatching scheme is passed;

and if the verification of the initial power dispatching scheme is passed, taking the initial power dispatching scheme as the power dispatching scheme of the power distribution network.

In a second aspect, the present invention further provides a power grid energy storage scheduling system based on deep learning and energy management, which includes a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor implements the power grid energy storage scheduling method based on deep learning and energy management as described in the first aspect when executing the computer program.

The beneficial effects of the invention are as follows:

By utilizing advanced machine learning algorithms such as a deep convolutional neural network, a support vector machine and the like, the invention can more accurately predict the power consumption load and the power generation amount of the micro-grid, and the high-precision prediction capability greatly improves the accuracy and reliability of power grid dispatching and avoids the problem of electric energy waste or shortage caused by inaccurate prediction. In addition, the invention fully considers the multidimensional information such as the historical electricity utilization data, the real-time electricity utilization data, the historical weather data, the weather forecast data and the like, and can comprehensively know and forecast the electricity utilization requirement and the electricity generation capacity of the distribution area by comprehensively analyzing the data, thereby realizing more scientific and reasonable electric energy dispatching. The invention can also utilize the residual stored electric energy data of the shared energy storage device, can effectively balance the supply and demand of the power grid, and avoid the problems of electric energy waste or unstable power grid caused by insufficient or excessive power generation of a single micro-power grid.

On the other hand, the invention realizes the data sharing and collaborative scheduling among all energy storage devices in the power distribution network by constructing the energy scheduling blockchain network, and the introduction of the blockchain technology not only ensures the safety and the transparency of the data, but also improves the execution efficiency and the reliability of a scheduling scheme. In the distributed scheduling mode, the energy storage device of each power distribution area can flexibly adjust own energy storage and discharge strategies according to real-time electric energy requirements and power generation conditions, so that the optimal scheduling of the whole power grid system is realized. The flexible and intelligent scheduling mode not only remarkably improves the utilization efficiency of electric energy and reduces the energy waste, but also enhances the stability and the risk resistance of the power grid system.

Drawings

Fig. 1 is a schematic flow chart of a power grid energy storage scheduling method based on deep learning and energy management in one embodiment of the application.

Fig. 2 is a schematic flow chart of a predicted electricity utilization characteristic of a predicted electricity utilization load in one embodiment of the application.

Fig. 3 is a schematic flow chart of predicting surplus power of a micro grid according to an embodiment of the application.

Detailed Description

The technical solutions of the embodiments of the present application will be clearly described below with reference to the drawings in the embodiments of the present application, and it is apparent that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which are obtained by a person skilled in the art based on the embodiments of the present application, fall within the scope of protection of the present application.

The terms first, second and the like in the description and in the claims, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged, as appropriate, such that embodiments of the present application may be implemented in sequences other than those illustrated or described herein, and that the objects identified by "first," "second," etc. are generally of a type, and are not limited to the number of objects, such as the first object may be one or more. Furthermore, in the description and claims, "and/or" means at least one of the connected objects, and the character "/", generally means that the associated object is an "or" relationship.

Referring to fig. 1, fig. 1 is a schematic flow chart of a power grid energy storage scheduling method based on deep learning and energy management in one embodiment, the power grid energy storage scheduling method based on deep learning and energy management shown in fig. 1 is applied to a power distribution network, the power distribution network includes a plurality of power distribution areas, each power distribution area includes a plurality of power loads and a plurality of micro-grids constructed based on a photovoltaic wind power generation system, and each power distribution area is further configured with a shared energy storage device, the shared energy storage device is used for storing surplus electric energy produced by all micro-grids in the power distribution area, and the surplus electric energy represents partial electric energy of the electric energy produced by the micro-grids exceeding the internal energy supply of the micro-grids in the same time period.

It should be understood that, although the steps in the flowchart of fig. 1 are shown in sequence as indicated by the arrows, the steps are not necessarily performed in sequence as indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in fig. 1 may include multiple sub-steps or stages that are not necessarily performed at the same time, but may be performed at different times, nor do the order in which the sub-steps or stages are performed necessarily performed in sequence, but may be performed alternately or alternately with at least a portion of other steps or sub-steps of other steps. As shown in fig. 1, the power grid energy storage scheduling method based on deep learning and energy management disclosed by the invention specifically comprises the following steps:

S101, acquiring historical electricity utilization data and real-time electricity utilization data of all electricity utilization loads in each power distribution area.

And collecting historical electricity utilization data of all the electricity utilization loads. The historical electricity consumption data generally comprises information of electricity consumption quantity, electricity consumption time, electricity consumption mode and the like of each electric equipment. The data can be obtained from various channels such as intelligent ammeter, electricity recording system, sensor of electric equipment and the like. Smart meters are the most common source of data that can record the power usage of each consumer in real time and transmit the data to a central database. The electricity consumption recording system can provide more detailed electricity consumption history data, including the change trend of electricity consumption, electricity consumption peak time period and the like. The sensors of the electric equipment can monitor the running state and the electricity consumption condition of the equipment, and more details are provided for data collection. In addition to historical electricity usage data, real-time electricity usage data is also required. The real-time electricity consumption data refers to electricity consumption conditions of the electric equipment at the current moment, and comprise real-time electricity consumption, equipment running states and the like. These data may be obtained in real time by smart meters and sensors and transmitted to a data center via a wireless or wired network. To ensure accuracy and integrity of the data, the collected data needs to be cleaned and preprocessed. The data cleaning aims to remove missing values and abnormal values and ensure the integrity and accuracy of data. The missing values can be processed by interpolation or filling, and the abnormal values need to be identified and removed according to a statistical method or business rules. The data preprocessing further comprises data normalization processing, wherein data with different dimensions are converted into the same numerical range, and dimensional differences among different features are eliminated.

S102, constructing an electricity utilization characteristic recognition model based on the deep convolutional neural network and through historical electricity utilization data, analyzing real-time electricity utilization data by utilizing the electricity utilization characteristic recognition model, and predicting to obtain predicted electricity utilization characteristics of electricity utilization loads.

In the electricity utilization characteristic identification, historical electricity utilization data can be regarded as time sequence data, and the electricity utilization quantity of each electric equipment changes with time. The first step in constructing the electrical signature model is data preparation. And arranging the collected historical electricity utilization data according to a time sequence to form a time sequence data set. The data set is then divided into a training set for training the model and a test set for evaluating the performance of the model. Next, the architecture of the deep convolutional neural network is designed. Convolutional neural networks typically include multiple convolutional layers, pooling layers, and fully-connected layers. The convolution layer is used for extracting local features in the data, the pooling layer is used for dimension reduction and feature selection, and the full connection layer is used for classification and regression. In power usage feature recognition, a network architecture including multiple convolutional layers and pooling layers may be designed to extract timing features and patterns in the power usage data. The model is then trained using the training set. During training, the model continuously adjusts parameters through a back-propagation algorithm to minimize the prediction error. After training, the model is evaluated by using the test set, and the accuracy and the robustness of the model are verified. In this way, an electrical signature model can be obtained. By using the model, real-time electricity consumption data can be analyzed, and electricity consumption characteristics of an electricity consumption load can be predicted.

S103, calculating to obtain the predicted electricity consumption total amount of the power distribution area according to the predicted electricity consumption characteristics of all the electricity consumption loads.

Firstly, the predicted electricity utilization characteristics of each electricity utilization load are summarized, and an electricity utilization characteristic data set of the whole power distribution area is obtained. This data set includes information of the predicted power usage, power usage pattern, etc. of each power usage load over a period of time in the future. Then, based on these electricity usage characteristic data, a predicted total amount of electricity usage for the power distribution area is calculated. Specifically, the predicted power consumption of each power consumption load may be accumulated to obtain the total power consumption of the power distribution area. To improve the accuracy of the prediction, various methods may be used to correct and optimize the amount of electricity used. For example, the predicted power usage may be adjusted based on trends and seasonal changes in historical data. In this way, a more accurate prediction of the total amount of electricity used can be obtained.

S104, acquiring historical weather data and weather forecast data of a power distribution area and historical surplus electric energy data of all micro-grids.

Wherein the historical weather data and the historical surplus electrical energy data are in the same historical time period. Historical weather data typically includes meteorological elements such as temperature, humidity, wind speed, rainfall, and the like, and in addition to the historical weather data, weather forecast data needs to be obtained. Weather forecast data is a weather forecast over a future period of time generated by numerical modeling and statistical methods based on weather models and observations. Such data may be obtained from a weather prediction system. In order to ensure the accuracy and consistency of the data, the weather prediction data needs to be preprocessed, including the steps of data cleaning, interpolation, normalization and the like. In addition to weather data, historical surplus energy data for all micro-grids needs to be collected. Micro-grid refers to a small-sized power system composed of distributed energy sources, energy storage devices, loads and the like in a specific area. The historical surplus electric energy data of the micro-grids comprise information such as generated energy, used energy and surplus electric energy of each micro-grid in different time periods. These data may be obtained from the monitoring system of the micro-grid, the energy management system, etc. In order to ensure the integrity and consistency of the data, the historical surplus electric energy data needs to be cleaned and preprocessed, including the steps of removing missing values and abnormal values, normalizing the data and the like.

S105, a power generation prediction model is built based on the support vector machine and through historical weather data and historical surplus power data, the weather prediction data are preprocessed, the preprocessed weather prediction data are input into the power generation prediction model, and the predicted surplus power of the micro-grid is obtained.

In the power generation prediction, the historical weather data and the historical surplus electric energy data can be regarded as characteristics and labels, and a power generation prediction model can be constructed through a support vector machine. The first step in constructing a power generation predictive model is data preparation. And arranging the collected historical weather data and the historical surplus electric energy data according to a time sequence to form a time sequence data set. The data set is then divided into a training set for training the model and a test set for evaluating the performance of the model. Next, a model structure of the support vector machine is designed. The support vector machine classifies data into different categories by finding an optimal hyperplane. In the power generation prediction, historical weather data can be used as characteristics, historical surplus electric energy data is used as a label, and a power generation prediction model is constructed by learning the relation between the characteristics and the label through a support vector machine. The model is then trained using the training set. In the training process, the model continuously adjusts parameters through an optimization algorithm to minimize the prediction error. After training, the model is evaluated by using the test set, and the accuracy and the robustness of the model are verified. In this way, a power generation prediction model can be obtained. By using the model, weather prediction data can be analyzed, and surplus electric energy of the micro-grid can be predicted.

S106, calculating to obtain the predicted energy storage data of the shared energy storage device by combining the predicted electricity consumption total amount, the predicted surplus electric energy of all the micro-grids and the residual stored electric energy of the shared energy storage device, and storing the predicted energy storage data in an energy storage server carried by the shared energy storage device.

The total predicted power consumption amount is subtracted from the sum of the predicted surplus power and the residual stored power, so that the predicted energy storage data of the shared energy storage device can be simply calculated. After calculation is completed, the predicted energy storage data is stored in an energy storage server carried by the shared energy storage device, and the energy storage server is a cloud server specially used for storing and managing the energy storage data.

S107, an energy scheduling block chain network of the power distribution network is built based on the energy storage server of each shared energy storage device, and an electric energy scheduling scheme of the power distribution network is generated by combining all the predicted energy storage data and through the energy scheduling block chain network.

The blockchain is a decentralised distributed account book technology, and ensures the security and the non-tamper property of data through an encryption algorithm and a consensus mechanism. In energy scheduling, a decentralized energy scheduling network can be constructed by using a blockchain technology, and an energy storage server of each shared energy storage device is used as a node of the blockchain network. Each node exchanges and shares data through a block chain protocol, and the safety and consistency of the data are ensured. The first step in building an energy scheduling blockchain network is node deployment. And deploying the energy storage server of each shared energy storage device as a block chain node, and carrying out data transmission and consensus among the nodes through an encryption communication protocol.

Next, consensus mechanisms and smart contracts for blockchain networks are designed. Common consensus mechanisms are used to ensure consistency and security of data in blockchain networks, including workload certification, rights certification, and the like. An intelligent contract is an automated program running on a blockchain for executing and validating transactions. In energy scheduling, intelligent contracts may be designed to automatically perform the generation and execution of power scheduling schemes. All the predicted stored energy data is then uploaded into the blockchain network. Each node packages the predicted energy storage data of the node into blocks, and verifies and records the blocks through a consensus mechanism. In this way, security and tamper-resistance of the data can be ensured. Next, a power scheduling scheme for the power distribution network is generated by the smart contract. The intelligent contract automatically calculates and optimizes the power scheduling scheme based on the predicted energy storage data and the energy demand. Specifically, distribution and scheduling of electric energy can be reasonably arranged according to electricity demand, generated energy and stored energy, and stable and efficient power supply of a power distribution area is ensured.

In one embodiment, referring to fig. 2, a power consumption feature recognition model is constructed based on a deep convolutional neural network and through historical power consumption data, real-time power consumption data is analyzed by using the power consumption feature recognition model, and predicted power consumption features of a predicted power consumption load are obtained by prediction, including the following steps:

S201, constructing an initial electricity utilization characteristic recognition model based on a deep convolutional neural network.

S202, preprocessing all historical electricity utilization data into a historical electricity utilization data set.

All the historical electricity utilization data are marked with corresponding historical electricity utilization characteristics in advance.

S203, performing feature correlation analysis on the historical electricity utilization features of the historical electricity utilization data set to obtain an electricity utilization feature classification matrix of the historical electricity utilization data set.

S204, training an initial electricity utilization characteristic recognition model by using the electricity utilization characteristic classification matrix to obtain a trained electricity utilization characteristic recognition model.

S205, analyzing real-time electricity utilization data by using an electricity utilization characteristic recognition model, and predicting to obtain predicted electricity utilization characteristics of an electricity utilization load.

In this embodiment, the Deep Convolutional Neural Network (DCNN) is a powerful tool applied to time series data analysis, and the multi-layer structure of the DCNN can effectively extract and learn complex features in data. In the electricity utilization feature recognition, the DCNN has the advantage that feature information in electricity utilization data can be extracted layer by layer through hierarchical convolution operation, so that accurate electricity utilization feature classification and recognition are realized. First, a hierarchy of networks needs to be defined, including an input layer, multiple convolution layers, a pooling layer, a fully connected layer, and an output layer. The input layer is responsible for receiving power usage data, which may include timing data for voltage, current, power, etc. The convolution layer extracts local features through convolution operation, and the size and the number of convolution kernels are super-parameters which need to be adjusted in an important way. The pooling layer reduces the data dimension by downsampling while preserving important features, common pooling operations include maximum pooling and average pooling. The full connection layer synthesizes the extracted features, and the output layer gives out a final electricity utilization feature recognition result.

To improve the performance of the model, a batch normalization layer may be added after the convolution layer, which helps to speed up the training process and stabilize the training effect of the model. Activating a function, such as ReLU (modified linear units), is used to enhance the nonlinear expression capabilities of the model so that the model can fit complex patterns of electrical characteristics. The training process of the model uses large-scale historical electricity utilization data, and network parameters are adjusted through a back propagation algorithm, so that the model can accurately identify and classify electricity utilization characteristics. The back propagation algorithm directs the direction and magnitude of the update of the parameters by calculating the gradient of the loss function, and common optimization algorithms include random gradient descent (SGD) and its variants such as Adam optimization algorithm. In the training process, super parameters such as learning rate, batch size, network layer number and the like need to be continuously adjusted to find the optimal model configuration. Furthermore, to avoid overfitting, regularization techniques such as L2 regularization or Dropout can be employed to randomly discard some of the neuron outputs, enhancing the generalization ability of the model. Through repeated iterative training and verification, the electricity utilization characteristic recognition model with excellent performance is finally obtained, and the model can accurately recognize and classify the characteristics in the electricity utilization data.

After the model is built, all historical electricity usage data needs to be preprocessed into a historical electricity usage data set. The purpose of the preprocessing is to clean up the data, eliminate noise and normalize the data to improve the effectiveness and accuracy of model training. First, all historical electricity usage data is collected, which typically includes information such as voltage, current, power, time stamps, etc. Next, the data is cleaned to remove missing values and abnormal values. The missing values can be processed by interpolation or filling, and the abnormal values need to be identified and removed according to a statistical method or business rules. The data is then normalized to convert the data of different dimensions to the same numerical range, for example normalizing the data to between 0 and 1. This step helps to eliminate dimensional differences between different features and avoids excessive impact of certain features on the model during training.

Each piece of historical electricity utilization data is pre-marked with corresponding historical electricity utilization characteristics. These characteristics may be the load type of the power consuming load, the historical power usage pattern, etc. The marking process can be completed through an automatic marking tool, so that each piece of data is ensured to have an explicit characteristic label. To further improve the quality and consistency of the data, data enhancement techniques may be employed to generate more training samples by randomly transforming the raw data, such as translating, rotating, scaling, etc. The data enhancement can not only increase the data quantity, but also improve the robustness of the model, so that the model can be better adapted to different power utilization scenes and characteristic changes. Finally, the historical electricity utilization data set after pretreatment is used as basic data of model training, and the data not only contains rich information of electricity utilization characteristics, but also is subjected to strict cleaning and marking, so that high quality and high reliability of the data are ensured.

The purpose of the feature correlation analysis is to identify and quantify the relationships between different electrical characteristics, thereby constructing an electrical characteristic classification matrix. First, a statistical analysis is required for each feature in the dataset to calculate its basic statistics. Then, correlation between the respective features is calculated using a correlation analysis method such as pearson correlation coefficient, spearman correlation coefficient, or the like. Correlation analysis can help identify those features that have a greater impact on the power usage pattern and exclude redundant or irrelevant features. Then, the high-dimensional characteristic data is reduced to a low-dimensional space by using a dimension reduction technology such as Principal Component Analysis (PCA) or Factor Analysis (FA) so as to facilitate subsequent classification and model training. Through the analysis methods, an electricity utilization characteristic classification matrix can be constructed, and the matrix reflects the association relation and importance among different electricity utilization characteristics. The construction of the feature classification matrix is beneficial to improving the training efficiency and accuracy of the model, avoiding the model from being interfered by redundant features in the training process, and enhancing the generalization capability of the model.

Training the initial electricity utilization feature recognition model by using the electricity utilization feature classification matrix is a key step. Specifically, the preprocessed historical electricity usage dataset and the electricity usage feature classification matrix may be used as input data for the model. The data set is divided into a training set and a verification set according to a certain proportion, the training set is used for adjusting and optimizing model parameters, and the verification set is used for evaluating the performance of the model. In the model training process, a cross entropy loss function or a mean square error loss function is used for measuring the difference between the prediction result of the model and the actual label. The gradient of the model parameters is calculated by a back-propagation algorithm and the model parameters are updated using a gradient descent algorithm or a variation thereof, such as Adam optimization algorithm. To avoid overfitting, regularization techniques such as L2 regularization or Dropout can be employed to randomly discard some of the neuron outputs, enhancing the generalization ability of the model. In the training process, super parameters such as learning rate, batch size, network layer number and the like need to be continuously adjusted to find the optimal model configuration.

Through repeated iterative training and verification, the electricity utilization characteristic recognition model with excellent performance is finally obtained, and the model can accurately recognize and classify the characteristics in the electricity utilization data. In the training process, a cross-validation method can be adopted to divide the data set into a plurality of subsets, one subset is used as a validation set in turn, and the other subsets are used as training sets, so that the stability and generalization capability of the model under different data distribution are ensured. In addition, in order to further improve the performance of the model, an integrated learning method can be adopted to train a plurality of different models, and the prediction results of the models are fused to obtain more accurate and robust prediction results. The integrated learning can be realized in various modes, such as voting method, weighted average method and the like, different weights are given according to the performances of different models, the advantages of each model are integrated, and the accuracy and stability of the final prediction result are improved.

And analyzing and predicting the real-time electricity utilization data by using the trained electricity utilization characteristic recognition model. Firstly, preprocessing the power consumption data acquired in real time, wherein the steps are similar to preprocessing of historical data, including data cleaning, normalization and the like. And then, inputting the preprocessed real-time electricity utilization data into an electricity utilization feature recognition model, performing layer-by-layer processing on the input data by the model through a forward propagation process, extracting features, classifying or regressing, and outputting a prediction result. The prediction result can be the type of the electricity load, the electricity mode and other characteristic information. In order to verify the accuracy of the prediction result, the prediction result can be compared with actual electricity consumption data, and error indexes such as mean square error, absolute error and the like are calculated. If the error is within an acceptable range, it indicates that the predictive performance of the model is good. Otherwise, further optimization of model parameters or adjustment of model structure is required.

In one embodiment, performing feature correlation analysis on historical electricity utilization features of a historical electricity utilization data set to obtain an electricity utilization feature classification matrix of the historical electricity utilization data set includes the following steps:

extracting historical electricity utilization characteristics of a historical electricity utilization data set;

Calculating by adopting a connection function to obtain the relevance of the power utilization characteristics of the historical power utilization characteristics of different characteristic types;

And analyzing the relevance of the power utilization characteristics by utilizing a decision tree learning algorithm based on the information entropy, and acquiring a power utilization characteristic classification matrix of the historical power utilization data set.

In this embodiment, after the historical electricity usage characteristics are extracted, the next step is to calculate the degree of association between the electricity usage characteristics of each different characteristic type. The calculation of the association degree is helpful for understanding the relation between different features, and provides reference for subsequent feature selection and model optimization. The use of a join function is a common method of calculating the degree of association. First, an appropriate connection function is selected. Common connection functions include pearson correlation coefficients, spearman rank correlation coefficients, mutual information, and the like. The pearson correlation coefficient is used to measure the linear relationship between two variables, and has a value ranging from [ -1,1], where 1 represents a complete positive correlation, -1 represents a complete negative correlation, and 0 represents no correlation. The spearman rank correlation coefficient is used for measuring the monotonic relation between two variables and is suitable for nonlinear relation. Mutual information is used to measure the dependency between two variables, and is not limited by linear or nonlinear relationships.

In the implementation process, firstly, all feature vectors need to be standardized to eliminate the influence among different feature dimensions. The normalization method may select a mean-variance normalization or a min-max normalization. The mean-variance normalization divides the standard deviation by subtracting the mean from the eigenvalue, so that the normalized eigenvalue is 0 in the mean and 1 in the standard deviation. The min-max normalization then scales the eigenvalues to within the [0,1] range. Taking the spearman rank correlation coefficient as an example, the eigenvalues first need to be converted into ranks, and then the correlation coefficients between ranks are calculated. The calculation formula is as follows:

,

Where di represents the rank difference between the two historical electricity usage characteristics and n represents the number of historical electricity usage characteristics.

After the relevance of the electricity utilization characteristics is calculated, the relevance is analyzed by utilizing a decision tree learning algorithm based on information entropy, and an electricity utilization characteristic classification matrix of the historical electricity utilization data set is obtained. Decision trees are a common classification and regression algorithm that builds a tree structure by recursively partitioning a data set into subsets. The information entropy is used to measure the uncertainty of the data set, and the decision tree selects the optimal segmentation point by maximizing the information gain. All historical electricity utilization features are built into an initial decision tree based on the degree of electricity utilization feature association. Each node of the decision tree represents a feature and each edge represents a segmentation condition of a feature value. The root node represents the entire dataset and the leaf nodes represent the classification results. In the process of constructing the decision tree, the information entropy of each feature is first calculated. The calculation formula of the information entropy is as follows:

,

Where D represents the historical electricity usage dataset and pi represents the probability of the i-th class of historical electricity usage feature. The information gain represents the reduction of the information entropy after the feature segmentation, and the calculation formula is as follows:

,

wherein A represents the historical electricity utilization characteristic, dv represents the subset of the historical electricity utilization characteristic A with the value v.

In the implementation process, firstly, the information gain of each feature is calculated, and the feature with the largest information gain is selected as a segmentation node. The dataset is then partitioned into subsets according to the values of the features, and a sub-decision tree is recursively built for each subset until all features are used or the samples in the subset belong to the same class. After the decision tree is constructed, pruning can be carried out on the decision tree according to the feature association degree. The purpose of pruning is to remove redundant nodes and improve the generalization capability of the model. Common pruning methods include pre-pruning and post-pruning. The pre-pruning is to limit the growth of the tree by setting the maximum depth, minimum sample number and other parameters in the process of constructing the decision tree, and the post-pruning is to remove unnecessary nodes by cross validation and other methods after the decision tree is constructed. Finally, the association degree of the electricity utilization characteristics is analyzed by utilizing the decision tree, and an electricity utilization characteristic classification matrix is obtained. Each row of the classification matrix represents a feature, each column represents a classification result, and the values in the matrix represent the distribution of the features under different classification results. The contribution of different features in the power utilization feature classification can be intuitively known through the classification matrix, and a basis is provided for subsequent feature selection and model optimization.

In one embodiment, training an initial electrical feature recognition model using an electrical feature classification matrix to obtain a trained electrical feature recognition model includes the steps of:

Constructing an optimal feature matrix based on the optimal electricity utilization feature by a weighted least square method;

Inputting the optimal feature matrix into an initial electricity utilization feature recognition model, and calculating to obtain bias item parameters of all the full-connection layers in the initial electricity utilization feature recognition model;

And if the parameter error is smaller than the error threshold, completing a model training process of the initial electricity utilization characteristic recognition model to obtain a trained electricity utilization characteristic recognition model.

In this embodiment, first, an optimal electricity utilization feature is selected from the electricity utilization feature classification matrix by using a particle swarm optimization algorithm. After the optimal power usage characteristics are determined, an optimal characteristics matrix is then required to be constructed by a weighted least Squares method (WEIGHTED LEAST square, WLS). The weighted least squares method is a linear regression method that improves the accuracy of the model fit by assigning different weights to different optimal power usage characteristics to minimize the sum of squares of the errors. In practice, it is first necessary to construct a design matrix X, where each row represents a sample and each column represents a feature. Then, a weight matrix W, which is a diagonal matrix in which elements on the diagonal represent the weight of each sample, is constructed. The optimal feature matrix can be obtained by a weighted least square methodThe calculation formula is as follows:

,

The optimal feature matrix constructed by the method can reflect the relation between the electricity consumption features and the electricity consumption to the maximum extent, thereby providing a solid foundation for subsequent model training.

And inputting the optimal feature matrix into an initial electricity utilization feature recognition model, carrying out weighted summation and nonlinear transformation on each layer by the optimal feature matrix through forward propagation, and finally obtaining a prediction result through an output layer. To calculate the bias term parameters for each layer, back propagation and gradient descent are required. The back propagation updates the weights and bias terms by calculating the partial derivatives of the loss function (e.g., mean square error) for each layer parameter. The specific formula is as follows:

,

Wherein: is the bias term after the t+1st iteration, Is the bias term after the t-th iteration,Is the learning rate and L is the loss function.

After the bias term parameters of all the connection layers in the initial power utilization characteristic identification model are obtained through calculation, parameter errors of the bias term parameters need to be calculated through preset optimal parameters. The parameter error refers to the difference between the actual calculated bias term parameter and the expected optimal parameter. This difference can be measured in a number of ways, most commonly using mean square Error (Mean Squared Error, MSE) or Absolute Error (Absolute Error). In the implementation, first, a preset optimal parameter needs to be defined. These parameters are typically derived through a priori knowledge or empirical data, representing the parameter values that the model should have under ideal conditions. Then, the error between the actual bias term parameter and the preset optimal parameter is calculated. If the error is greater than or equal to the preset error threshold, all the steps are needed to be re-executed, including reconstructing the optimal feature matrix, inputting the initial power utilization feature recognition model, calculating the bias term parameters of the connecting layer and the like until the errors of all the parameters are smaller than the error threshold.

Through the repeated iteration process, the model parameters can be continuously optimized, the parameter errors are gradually reduced, and finally the expected precision requirement is achieved. If the errors of all the parameters are smaller than the error threshold, the model training process is completed, and the trained electricity utilization characteristic recognition model is obtained.

In one embodiment, selecting the optimal electricity utilization feature from the electricity utilization feature classification matrix by using a particle swarm optimization algorithm comprises the following steps:

initializing initial electricity utilization characteristics into two-dimensional characteristic data by using preset particle swarm algorithm parameters, and constructing initial particle swarms according to the two-dimensional characteristic data;

And taking all initial electricity utilization characteristics in the optimal particle swarm as optimal electricity utilization characteristics.

In this embodiment, after the electricity feature classification matrix is obtained, initial electricity features are randomly selected from the electricity feature classification matrix, and an optimal electricity feature is selected from the initial electricity features by using a Particle Swarm Optimization (PSO) algorithm. PSO is an optimization algorithm simulating group intelligence, and optimal solutions are found by simulating the foraging behavior of the bird group. Each particle represents a candidate solution and the solution space is searched by continuously updating the position and velocity. Specifically, firstly, initializing initial electricity utilization characteristics into two-dimensional characteristic data by using preset particle swarm algorithm parameters, and constructing initial particle swarms according to the two-dimensional characteristic data. Each particle in the initial population of particles represents a feature subset, the dimension of the particle being equal to the number of two-dimensional feature data. The initial position and speed of the particles can be randomly generated, and can also be initialized according to the feature importance of the feature classification matrix. Each particle has an fitness value that is used to measure the merit of the feature subset. The fitness value may be calculated based on the classification accuracy, information gain, etc. metrics.

In a specific implementation process, the size of the particle swarm and the iteration number are first determined. The size of the particle swarm determines the coverage of the search space, and the iteration number determines the termination condition of the algorithm. Generally, larger particle populations and more iterations can increase search accuracy, but also increase computational complexity. The velocity and position of the particles are then updated. The velocity update formula for the particles is:

,

where vi (t) denotes the velocity of particle i at the t-th generation, xi (t) denotes the position of particle i at the t-th generation, pi denotes the historical optimal position of particle i, pg denotes the global optimal position, w denotes the inertial weight, c1 and c2 denote acceleration constants, and r1 and r2 are random numbers between [0,1 ].

The location update formula of the particles is:

,

Through the above formula, each particle is continuously moved in the search space, gradually approaching the optimal solution. In each iteration, the fitness value of each particle is updated and the global optimal position is recorded. When the maximum iteration number or fitness value is reached and no longer significantly improved, the algorithm terminates and returns the global optimal position as the optimal feature subset. Through a PSO algorithm, optimal electricity utilization characteristics are selected from the electricity utilization characteristic classification matrix, so that the accuracy and efficiency of characteristic selection can be effectively improved, and a high-quality characteristic set is provided for subsequent model training and prediction.

In one embodiment, referring to fig. 3, a power generation prediction model is constructed based on a support vector machine and through historical weather data and historical surplus power data, the weather prediction data is preprocessed, the preprocessed weather prediction data is input into the power generation prediction model, and the predicted surplus power of the micro-grid is obtained, which comprises the following steps:

s301, constructing an initial power generation prediction model based on a support vector machine.

S302, respectively calculating the data association degree between the historical weather data and the historical surplus electric energy data of different types by using a gray association algorithm.

S303, integrating all the historical weather data with the data association degree larger than a preset association degree threshold value into a historical weather data set.

S304, carrying out iterative training on the initial power generation prediction model by combining the historical weather data set and the historical surplus electric energy data until model parameters in the initial power generation prediction model are trained to optimal model parameters, and obtaining a power generation prediction model after training.

S305, preprocessing the weather prediction data, and inputting the preprocessed weather prediction data into a power generation prediction model to obtain the predicted surplus electric energy of the micro-grid.

In this embodiment, the support vector machine is a supervised learning model, and is widely used for classification and regression analysis. The core idea is to divide the data by finding an optimal hyperplane in the high-dimensional space, thereby realizing the prediction of new data. After the initial power generation prediction model is constructed, the relationship between the historical weather data and the historical surplus power data needs to be analyzed next. The gray correlation algorithm is an effective analysis tool that can quantify the degree of correlation between different variables. The historical weather data can be paired with corresponding historical surplus electrical energy data to form a multi-dimensional dataset. The gray correlation algorithm measures the degree of correlation of variables by calculating the gray correlation degree between the variables. Specifically, the gray correlation is determined by calculating the degree of relative change between the variables. First, it is necessary to perform dimensionless processing on data so that data of different dimensions can be compared. Then, a difference sequence between each variable and a reference variable (historical surplus electric energy data) is calculated, and gray correlation is obtained by solving the absolute value and the maximum and minimum values of the difference sequence. The calculated gray correlation degree can reflect the correlation degree between each weather variable and the historical surplus electric energy data. The higher the correlation, the greater the impact of the weather variable on surplus power. By using the method, key weather factors such as temperature, humidity, wind speed and the like can be screened out, and the key weather factors can directly influence the power generation efficiency of the photovoltaic wind power generation, so that the influence on surplus electric energy is larger.

After gray correlation between each weather variable and the historical surplus energy data is calculated, the next step is to screen out weather data with higher correlation. A correlation threshold is set, and only weather variables having a correlation greater than the threshold are included in the historical weather dataset. The setting of this threshold can be adjusted according to practical situations and experience, and a value is usually selected that can significantly distinguish between a high degree of association and a low degree of association. And integrating the screened weather variables with high correlation degree into a new historical weather data set. This data set contains critical weather factors that have a greater impact on surplus power. In this way, the redundancy of the data can be reduced, and the quality of the data and the training efficiency of the model can be improved.

After screening out the historical weather data with high correlation, combining the data with the historical surplus electric energy data, and performing iterative training of an initial power generation prediction model. Firstly, the integrated historical weather data set and the historical surplus electric energy data are paired to form a new training data set. This dataset will be used for iterative training of the initial power generation predictive model to optimize model parameters. The iterative training process is a process of continuously adjusting and optimizing model parameters. First, a new training data set is input into an initial power generation prediction model, and training is performed by a support vector machine algorithm. During training, the model continuously adjusts parameters according to the input data to minimize the prediction error. This process typically requires multiple iterations, each of which adjusts the model parameters and calculates new prediction errors.

In order to improve the training efficiency and the prediction accuracy of the model, a cross-validation method can be adopted. The training data set is divided into a plurality of subsets, one subset being used as the validation set at a time, the remaining subsets being used as training sets. By the method, the performance of the model can be more comprehensively estimated, and the problems of over fitting and under fitting are avoided. After each iterative training, the prediction performance of the model is evaluated, and the prediction error and the evaluation index are calculated. By analyzing these indices, the degree of optimization of the model parameters can be determined. And stopping iterative training if the prediction performance of the model reaches the expected target, and obtaining a final power generation prediction model. Otherwise, continuing to adjust the model parameters, and carrying out the next iteration training until the model performance reaches the optimal. After the trained power generation prediction model is obtained, the model can be used for actual prediction.

In one embodiment, preprocessing weather prediction data, inputting the preprocessed weather prediction data into a power generation prediction model, and obtaining predicted surplus power of the micro-grid includes the following steps:

extracting all weather data types in the historical weather data set;

screening out invalid weather prediction data with different data types from all weather prediction data types;

And inputting all the residual weather prediction data into a power generation prediction model to obtain the predicted surplus electric energy of the micro-grid.

In this embodiment, the consolidated historical weather dataset is categorized and extracted using a data analysis tool or programming language. Specifically, a script may be written that traverses each record in the history weather data set, identifying and extracting the weather data type therein. After the historical weather data types are extracted, invalid weather prediction data having a different weather data type from all weather data types needs to be filtered out. The purpose of this step is to remove those invalid data that are not satisfactory, ensuring consistency and accuracy of the data input into the power generation prediction model. After the invalid weather prediction data is filtered out, the remaining data is retained as valid weather prediction data. These valid data will be used for subsequent power generation prediction model inputs. In this way, the predicted surplus energy of the micro grid can be predicted.

In one embodiment, the energy scheduling block chain network of the distribution network is constructed based on the energy storage server of each shared energy storage device, and the power scheduling scheme for combining all the predicted energy storage data and generating the distribution network through the energy scheduling block chain network comprises the following steps:

constructing an energy scheduling blockchain network of a power distribution network based on an energy storage server of each shared energy storage device;

Marking the shared energy storage device with the predicted energy storage data lower than a preset electric energy data threshold as a target shared energy storage device, and marking an energy storage server of the target shared energy storage device as a target energy storage server;

Automatically generating an electric energy scheduling request of the target shared energy storage device through the target energy storage server, and uploading the electric energy scheduling request to an energy scheduling block chain network;

Broadcasting the electric energy scheduling request to other energy storage servers except the target energy storage server by utilizing the leading server;

when the energy storage server receives the electric energy scheduling request, generating electric energy scheduling information by combining the electric energy scheduling request and the locally stored predicted energy storage data, and uploading the electric energy scheduling information to an energy scheduling block chain network;

In this embodiment, an energy scheduling blockchain network of the power distribution grid is built based on each energy storage server sharing energy storage devices. The core of the step is to realize efficient management and scheduling of the shared energy storage device by using the distributed ledger and intelligent contract functions of the blockchain technology. In carrying out this step, it is first necessary to configure each shared energy storage device with an energy storage server that is responsible for managing and monitoring the status and data of the respective energy storage device. Next, the energy storage servers are connected using blockchain technology to construct a decentralized energy scheduling network. The distributed ledger feature of blockchain technology can ensure data sharing and synchronization between all servers, while its non-tamper ability can ensure security and trustworthiness of the data. In particular, a suitable blockchain platform, such as an ethernet or HYPERLEDGER FABRIC, may be selected to deploy a private or federated chain. Each energy storage server will be added to the blockchain network as a node, and the consistency of the ledgers is maintained between the nodes through a consensus mechanism (such as PoW, poS or PBFT). The intelligent contracts are then used to automatically execute business logic associated with energy scheduling, such as generation and processing of power scheduling requests.

After the energy scheduling blockchain network is built, the next step is to elect the leader server in all the energy storage servers according to the server states of the energy storage servers. The leader server plays a role in coordination and management in the whole network, is responsible for processing and distributing the power scheduling request, and ensures the efficient operation of the system. In implementing this step, it is first necessary to define evaluation criteria for the server state, which may include computing power of the server, network delay, storage capacity, online time, etc. By comprehensively evaluating these criteria, the best performing server can be selected as the leader server. In particular, an off-centered election algorithm, such as Raft or Paxos, may be used that can elect a leader efficiently in a distributed system.

In the election process, each energy storage server can carry out self-evaluation according to the state information of the energy storage server, and the evaluation result is broadcast to other servers. After all the servers receive the information, voting is carried out according to a preset election rule, and finally the server with the largest vote is selected as a leading server. All operations and results during election are recorded on the blockchain, ensuring transparency and traceability. After the leader server is elected, the server is responsible for coordinating and managing the whole energy scheduling network, processing the electric energy scheduling request and distributing the electric energy scheduling request to other servers, and ensuring the efficient operation of the system and the reasonable distribution of electric energy.

After the leader server is elected, the shared energy storage device needing to be subjected to electric energy scheduling is identified according to the predicted energy storage data. Specifically, the shared energy storage devices for which the predicted energy storage data is below a preset power data threshold are marked as target shared energy storage devices, and the energy storage servers of these devices are marked as target energy storage servers. In performing this step, it is first necessary to obtain predictive energy storage data for each shared energy storage device, which is typically predicted by the energy storage server based on historical data and current status. Next, a power data threshold is set, which may be set according to the system requirements and security requirements. For example, a minimum energy storage level may be set, and when the predicted energy storage data is below this level, the device may need to schedule power.

After the target shared energy storage device and the target energy storage server are marked, the next step is to automatically generate electric energy scheduling requests of the target shared energy storage device through the target energy storage server, and to uplink the requests to the energy scheduling blockchain network. The core of the step is to automatically generate and process the power scheduling request by using intelligent contracts and automation technology. In implementing this step, smart contracts are first deployed on the target energy storage server, which are used to define the generation rules and flows of the power scheduling request. The intelligent contract can automatically generate an electric energy scheduling request according to the predicted energy storage data and the current state of the target shared energy storage device. For example, when the predicted stored energy data is below a set threshold, the intelligent appointment triggers generation of a power scheduling request, specifying information such as power and time to be scheduled. The generated power schedule request will be recorded on the blockchain, ensuring the security and non-tamper ability of the data. Through the blockchain network, power scheduling requests will be broadcast to all relevant nodes, ensuring that all energy storage servers can receive and process these requests.

After generating and linking the power schedule requests, the next step is to broadcast the requests to other energy storage servers than the target energy storage server using the leader server. In doing so, the leader server periodically scans the blockchain for power scheduling requests, identifying requests that need to be broadcast. The leader server then broadcasts these requests to the other energy storage servers over the blockchain network. In the broadcasting process, the leader server can select the optimal broadcasting path according to the network topology structure and the node state, so that the request can be conveyed quickly. In particular, the power scheduling request may be sent to other servers via a messaging mechanism using a P2P communication protocol of a blockchain network. After each server receives the request, the server processes the request according to the content of the request and the state of the server, and feeds back the processing result to the leading server.

After the energy storage server receives the power scheduling request, the next step is to combine the request content with the locally stored predictive energy storage data to generate power scheduling information, and upload the information to the power scheduling blockchain network. The core of the step is to generate accurate power scheduling information by using local data and request content, so as to ensure the efficient operation of the system. When the step is implemented, the energy storage server firstly analyzes the received power dispatching request, and extracts dispatching information, such as power energy, time and the like, which need to be dispatched. The server then compares and analyzes these information with locally stored predictive stored energy data to generate specific power schedule information. This process may be implemented by writing a data processing script or invoking a smart contract. The generated power schedule information is recorded on the blockchain to ensure the security and non-tamper ability of the data. Through the blockchain network, the power scheduling information will be broadcast to all relevant nodes, ensuring that all energy storage servers can receive and process the information.

After all energy storage servers generate and upload the electric energy scheduling information, the final step is to collect the information through a leading server and generate an electric energy scheduling scheme of the distribution network. The key of the step is to integrate all scheduling information by utilizing the coordination and management functions of the leader server to generate a global power scheduling scheme, so that the efficient operation of the system and the reasonable distribution of power are ensured. In doing so, the leader server periodically scans the power schedule information on the blockchain and gathers all relevant schedule data. The leader server then aggregates and analyzes the data to generate a global power schedule. This process may be implemented by invoking a smart contract or writing a data processing script. The generated power schedule scheme will be recorded on the blockchain, ensuring the security and non-tamper ability of the data. Through the blockchain network, power scheduling schemes will be broadcast to all relevant nodes, ensuring that all energy storage servers can receive and execute these schemes.

In one embodiment, the power scheduling scheme for aggregating all power scheduling information and generating a power distribution network by a leader server includes the steps of:

Summarizing all the electric energy scheduling information through a leading server and generating an initial electric energy scheduling scheme of the power distribution network;

Broadcasting an initial power scheduling scheme to all energy storage servers through a leading server;

If the verification of the initial power dispatching scheme is not passed, acquiring power dispatching information regenerated by all energy storage servers through an energy dispatching block chain network, and repeatedly executing the scheme generation step and the scheme verification step until the verification of the initial power dispatching scheme is passed;

and if the verification of the initial power dispatching scheme is passed, taking the initial power dispatching scheme as a power dispatching scheme of the power distribution network.

In this embodiment, when the leader server gathers all the power scheduling information and generates an initial power scheduling scheme for the power distribution network, the leader server first collects the power scheduling information uploaded by all the energy storage servers from the blockchain network. The information includes a current energy storage state of each shared energy storage device, predicted energy storage data, and a previously generated power schedule request. The lead server comprehensively analyzes the data, and considers the energy storage requirement, the power supply capacity and the overall power balance of the system of each energy storage device. The leader server integrates the information by invoking an intelligent contract or a preset algorithm to generate a preliminary power scheduling scheme. The scheme aims at optimizing the distribution of electric energy, and ensures that each energy storage device can realize reasonable dispatching and use of the electric energy on the premise of meeting the minimum energy storage requirement.

After generating the initial power schedule scheme, the lead server broadcasts the scheme to all energy storage servers. The broadcasting process utilizes the P2P communication protocol of the block chain network to ensure that each energy storage server can timely receive an initial power scheduling scheme. After receiving the scheme, the energy storage server can carry out local verification on the scheme, and check whether the scheme accords with the energy storage state and the requirement of the energy storage server. Each step in the broadcast process is recorded on the blockchain to ensure the transparency and security of the data. In this way, the initial power scheduling scheme can be quickly communicated to each node in the overall network, ensuring that all energy storage servers can participate in the verification and execution of the scheme.

It is a critical step to verify whether the initial power scheduling scheme passes or not using a consensus mechanism. A consensus mechanism such as PoW, poS or PBFT ensures that all energy storage servers agree on an initial power scheduling scheme. After receiving the scheme, each energy storage server can verify according to the energy storage state and the requirement of the energy storage server, and the verification result is fed back to the block chain network. And the consensus mechanism gathers and analyzes all feedback results according to preset rules, and finally determines whether the initial power scheduling scheme passes verification. This process ensures the rationality and executable of the scheme while utilizing the distributed ledger characteristics of the blockchain to ensure the transparency and trustworthiness of the verification process.

If the verification of the initial power dispatching scheme is not passed, acquiring power dispatching information regenerated by all energy storage servers through an energy dispatching block chain network, and repeatedly executing the scheme generation step and the scheme verification step until the verification of the initial power dispatching scheme is passed. In this case, the energy storage server regenerates the power schedule information based on the new energy storage state and demand and uploads it to the blockchain network. The leader server gathers the information again, generates a new initial power scheduling scheme, and broadcasts the new initial power scheduling scheme to all the energy storage servers for verification. This looping process continues until all energy storage server solutions agree, ensuring that the final generated power scheduling solution can meet the overall requirements of the system and the specific requirements of each energy storage device.

And if the verification of the initial power dispatching scheme is passed, taking the initial power dispatching scheme as a power dispatching scheme of the power distribution network. In this case, the initial power scheduling scheme that passes verification is regarded as the final power scheduling scheme, and is recorded and distributed in the blockchain network. All energy storage servers perform electric energy scheduling operation according to the scheme, so that reasonable distribution and use of electric energy are ensured. By the mode, the whole power distribution networking system can realize efficient electric energy management and scheduling, and stable operation and reasonable electric energy distribution of each shared energy storage device are ensured. Meanwhile, the transparency and traceability of the whole process are ensured by using the distributed ledger and intelligent contract function of the blockchain technology, and powerful support is provided for optimizing and managing the system.

The invention also discloses a power grid energy storage scheduling system based on deep learning and energy management, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the power grid energy storage scheduling method based on deep learning and energy management described in any one of the embodiments is realized when the processor executes the computer program.

The processor may be a Central Processing Unit (CPU), or of course, according to actual use, other general purpose processors, digital Signal Processors (DSP), application Specific Integrated Circuits (ASIC), ready-made programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., and the general purpose processor may be a microprocessor or any conventional processor, etc., which is not limited in this respect.

The memory may be an internal storage unit of the computer device, for example, a hard disk or a memory of the computer device, or an external storage device of the computer device, for example, a plug-in hard disk, a Smart Memory Card (SMC), a secure digital card (SD), or a flash memory card (FC) provided on the computer device, or the like, and may be a combination of the internal storage unit of the computer device and the external storage device, where the memory is used to store a computer program and other programs and data required by the computer device, and the memory may also be used to temporarily store data that has been output or is to be output, which is not limited by the present application.

It will be appreciated by persons skilled in the art that the above discussion of any embodiment is merely exemplary and is not intended to imply that the scope of the application is limited to these examples, that combinations of technical features in the above embodiments or in different embodiments may also be implemented in any order, and that many other variations of the different aspects of one or more embodiments of the application as described above exist within the spirit of the application, which are not provided in detail for the sake of brevity.

One or more embodiments of the present application are intended to embrace all such alternatives, modifications and variations as fall within the broad scope of the present application. Accordingly, any omissions, modifications, equivalents, improvements and others which are within the spirit and principles of the one or more embodiments of the application are intended to be included within the scope of the application.

Claims

1. The utility model provides a power grid energy storage scheduling method based on deep learning and energy management, which is characterized in that the method is applied to the distribution network, the distribution network comprises a plurality of distribution areas, each distribution area comprises a plurality of electric loads and a plurality of micro-grids constructed based on a photovoltaic wind power generation system, each distribution area is also provided with a shared energy storage device, the shared energy storage device is used for storing surplus electric energy produced by all the micro-grids in the distribution area, the surplus electric energy represents partial electric energy of the electric energy produced by the micro-grids in the same time period beyond the internal energy supply of the micro-grids, and the method comprises the following steps:

an intelligent contract is deployed on the target energy storage server, the electric energy scheduling request of the target shared energy storage device is automatically generated through the target energy storage server, the electric energy scheduling request is uploaded to the energy scheduling blockchain network, the intelligent contract is used for defining the generation rule and flow of the electric energy scheduling request, and the intelligent contract is used for automatically generating the electric energy scheduling request according to the predicted energy storage data and the current state of the target shared energy storage device;

2. The deep learning and energy management-based power grid energy storage scheduling method according to claim 1, wherein the deep convolutional neural network-based power utilization feature recognition model is constructed through the historical power utilization data, the real-time power utilization data is analyzed by using the power utilization feature recognition model, and the predicted power utilization feature of the power utilization load is predicted and obtained by the method comprises the following steps:

3. The deep learning and energy management-based power grid energy storage scheduling method according to claim 2, wherein the performing feature correlation analysis on the historical power utilization features of the historical power utilization data set to obtain a power utilization feature classification matrix of the historical power utilization data set includes the following steps:

Extracting the historical electricity utilization characteristics of the historical electricity utilization data set;

4. The deep learning and energy management-based power grid energy storage scheduling method according to claim 3, wherein training the initial power utilization feature recognition model by using the power utilization feature classification matrix to obtain a trained power utilization feature recognition model comprises the following steps:

5. The deep learning and energy management-based power grid energy storage scheduling method according to claim 4, wherein the selecting the optimal power utilization feature from the power utilization feature classification matrix by using a particle swarm optimization algorithm comprises the following steps:

6. The deep learning and energy management-based power grid energy storage scheduling method according to claim 1, wherein the step of constructing a power generation prediction model based on a support vector machine and through the historical weather data and the historical surplus power data, preprocessing the weather prediction data, inputting the preprocessed weather prediction data into the power generation prediction model, and obtaining the predicted surplus power of the micro-grid comprises the following steps:

7. The deep learning and energy management-based power grid energy storage scheduling method according to claim 6, wherein the preprocessing the weather prediction data, inputting the preprocessed weather prediction data into the power generation prediction model, and obtaining the predicted surplus energy of the micro-grid comprises the following steps:

extracting all weather data types in the historical weather data set;

8. The deep learning and energy management based power grid energy storage scheduling method of claim 1, wherein the step of aggregating all the power scheduling information by the lead server and generating a power scheduling scheme for the power distribution network comprises the steps of:

9. A deep learning and energy management-based power grid energy storage scheduling system, comprising a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the deep learning and energy management-based power grid energy storage scheduling method according to any one of claims 1 to 8 is realized when the processor executes the computer program.