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

The grid energy storage dispatching method based on deep learning and energy management utilizes deep convolutional neural networks and support vector machines to predict electricity load and power generation. Combined with shared energy storage devices and blockchain networks, it solves the problem of traditional grid dispatching's difficulty in managing distributed new energy sources, and achieves efficient, reliable, and scientific power dispatching for the grid.

CN118822229BActive Publication Date: 2024-11-29WUHAN YICHEN CHUANGXIANG TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411308809.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-19
Publication Date
2024-11-29
Estimated Expiration
2044-09-19

AI Technical Summary

Technical Problem

Traditional power grid dispatching methods are difficult to effectively manage and dispatch distributed renewable energy sources, resulting in low power grid operating efficiency and serious energy waste.

Method used

A grid energy storage scheduling method based on deep learning and energy management is adopted. It predicts electricity load and power generation through deep convolutional neural networks and support vector machines, and combines shared energy storage devices and energy scheduling blockchain network to achieve efficient power scheduling.

Benefits of technology

It has improved the accuracy and reliability of power grid dispatch, reduced energy waste, enhanced the stability and resilience of the power grid, and achieved the scientific and rational allocation of electricity.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118822229B_ABST
    Figure CN118822229B_ABST
Patent Text Reader

Abstract

The present invention provides a power grid energy storage dispatching method and system based on deep learning and energy management, the method comprising the following steps: obtaining historical power consumption data and real-time power consumption data of power loads; constructing a power consumption feature recognition model based on a deep convolutional neural network, analyzing real-time power consumption data using the power consumption feature recognition model, and predicting the predicted power consumption characteristics of the power load; calculating the predicted total power consumption of the distribution area; constructing a power generation prediction model based on a support vector machine, 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 microgrid; calculating the predicted energy storage data of the shared energy storage device; constructing an energy dispatching blockchain network for distribution networking based on an energy storage server, combining all predicted energy storage data and generating an energy dispatching scheme for distribution networking through the energy dispatching blockchain network. The present invention has the effect of reasonably dispatching power grid energy storage.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of power grid energy storage scheduling, and specifically relates to a power grid energy storage scheduling method and system based on deep learning and energy management. Background Art

[0002] Traditional power grid systems are usually based on centralized power generation, with electricity transmitted from power plants to various power terminals. Therefore, the power dispatch of traditional power grid systems only needs to adopt a simple planning and dispatching scheme. With the development and application of renewable energy technologies, especially the rapid growth of distributed energy such as photovoltaic power generation and wind power generation, traditional power grid systems are facing huge changes and challenges.

[0003] Photovoltaic power generation and wind power generation are both extremely unstable and volatile, and are widely distributed. In order to avoid the waste of energy from photovoltaic power generation and wind power generation, energy storage devices are often configured for photovoltaic power generation areas and wind power generation areas to store the overflow of electricity in the corresponding areas. Since traditional planning and scheduling schemes only rely on power plant power generation data and a small amount of terminal power consumption data, traditional scheduling schemes have been unable to cope with complex and changeable power demand and supply situations, resulting in low grid operation efficiency and serious energy waste. Therefore, the concept of smart grid came into being. Smart grid aims to achieve efficient, reliable and intelligent operation of the power grid through information and communication technology and advanced control methods. In this context, how to effectively manage and schedule distributed energy and energy storage devices has become one of the core issues in the research and application of smart grids. Summary of the invention

[0004] The present invention provides a power grid energy storage scheduling method and system based on deep learning and energy management to solve the problems of low operating efficiency and serious energy waste of a power grid containing a large number of distributed new energy sources caused by traditional power grid scheduling methods.

[0005] In a first aspect, the present invention provides a power grid energy storage scheduling method based on deep learning and energy management, which is applied to a power distribution network, wherein the power distribution network includes multiple power distribution areas, each of which includes multiple power loads and multiple microgrids built based on photovoltaic and wind power generation systems, and each of which is also equipped with a shared energy storage device, which is used to store the surplus electric energy produced by all the microgrids in the distribution area, wherein the surplus electric energy represents the part of the electric energy produced by the microgrid in the same time period that exceeds the internal energy supply of the microgrid, and the method comprises the following steps:

[0006] For each of the power distribution areas, obtaining historical power consumption data and real-time power consumption data of all the power loads in the power distribution area;

[0007] Based on a deep convolutional neural network and using the historical power consumption data, a power consumption feature recognition model is constructed, and the real-time power consumption data is analyzed using the power consumption feature recognition model to predict the power consumption features of the power load;

[0008] Calculate the predicted total power consumption of the power distribution area according to the predicted power consumption characteristics of all the power loads;

[0009] Acquire historical weather data and weather forecast data of the power distribution area, and historical surplus electric energy data of all the microgrids, wherein the historical weather data and the historical surplus electric energy data are in the same historical time period;

[0010] Based on a support vector machine and using the historical weather data and the historical surplus electric energy data, a power generation prediction model is constructed, the weather prediction data is preprocessed, and the preprocessed weather prediction data is input into the power generation prediction model to obtain the predicted surplus electric energy of the microgrid;

[0011] Combine the predicted total power consumption, the predicted surplus power of all the microgrids and the remaining stored power of the shared energy storage device to calculate the predicted energy storage data of the shared energy storage device, and store the predicted energy storage data in the energy storage server carried by the shared energy storage device;

[0012] The energy storage server of each shared energy storage device builds an energy scheduling blockchain network for the distribution network, combines all the predicted energy storage data and generates an energy scheduling plan for the distribution network through the energy scheduling blockchain network.

[0013] Optionally, the step of constructing a power consumption feature recognition model based on a deep convolutional neural network and using the historical power consumption data, analyzing the real-time power consumption data using the power consumption feature recognition model, and predicting the predicted power consumption features of the power load comprises the following steps:

[0014] Construct an initial electricity consumption feature recognition model based on a deep convolutional neural network;

[0015] Preprocessing all the historical electricity consumption data into a historical electricity consumption data set, wherein all the historical electricity consumption data are pre-marked with corresponding historical electricity consumption features;

[0016] Performing feature correlation analysis on the historical power consumption features of the historical power consumption data set to obtain a power consumption feature classification matrix of the historical power consumption data set;

[0017] Using the power consumption feature classification matrix to train the initial power consumption feature recognition model to obtain a trained power consumption feature recognition model;

[0018] The real-time power consumption data is analyzed using the power consumption feature recognition model to predict the power consumption features of the power load.

[0019] Optionally, performing feature correlation analysis on the historical power consumption features of the historical power consumption data set to obtain a power consumption feature classification matrix of the historical power consumption data set includes the following steps:

[0020] Optionally, extracting the historical electricity consumption features of the historical electricity consumption dataset;

[0021] The connection function is used to calculate the power consumption feature correlation of the historical power consumption features of different feature types;

[0022] The power consumption feature correlation is analyzed using a decision tree learning algorithm based on information entropy, and a power consumption feature classification matrix of the historical power consumption data set is obtained.

[0023] Optionally, the using the power consumption feature classification matrix to train the initial power consumption feature recognition model to obtain a trained power consumption feature recognition model comprises the following steps:

[0024] Using a particle swarm optimization algorithm to select the optimal power consumption feature from the power consumption feature classification matrix;

[0025] Based on the optimal electricity consumption characteristics and using a weighted least squares method, an optimal characteristic matrix is ​​constructed;

[0026] Inputting the optimal feature matrix into the initial power consumption feature recognition model, and calculating the bias item parameters of all fully connected layers in the initial power consumption feature recognition model;

[0027] Calculating the parameter error of the bias term parameter by using the preset optimal parameter;

[0028] If the parameter error is greater than or equal to the preset error threshold, re-execute all the above steps until the parameter error is less than the error threshold;

[0029] If the parameter error is less than the error threshold, the model training process of the initial power consumption feature recognition model is completed to obtain a trained power consumption feature recognition model.

[0030] Optionally, the step of selecting the optimal power consumption feature from the power consumption feature classification matrix using a particle swarm optimization algorithm comprises the following steps:

[0031] Randomly selecting initial electricity usage features from the electricity usage feature classification matrix;

[0032] Initializing the initial power consumption characteristics into two-dimensional characteristic data using preset particle swarm algorithm parameters, and constructing an initial particle swarm according to the two-dimensional characteristic data;

[0033] The initial particle swarm is optimized multiple times by using a particle swarm optimization algorithm until the initial particle swarm is optimized to be an optimal particle swarm;

[0034] All the initial power usage characteristics in the optimal particle group are used as optimal power usage characteristics.

[0035] Optionally, the step of constructing a power generation prediction model based on a support vector machine and by using the historical weather data and the historical surplus electric 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 electric energy of the microgrid comprises the following steps:

[0036] Constructing an initial power generation prediction model based on support vector machine;

[0037] Using a grey correlation algorithm to respectively calculate the data correlation between different types of the historical weather data and the historical surplus electric energy data;

[0038] Integrate all the historical weather data whose data relevance is greater than a preset relevance threshold into a historical weather data set;

[0039] Iteratively training the initial power generation prediction model in combination with the historical weather data set and the historical surplus electric energy data until the model parameters in the initial power generation prediction model are trained to the optimal model parameters, thereby obtaining a trained power generation prediction model;

[0040] The weather forecast data is preprocessed, and the preprocessed weather forecast data is input into the power generation prediction model to obtain the predicted surplus electric energy of the microgrid.

[0041] Optionally, preprocessing the weather forecast data and inputting the preprocessed weather forecast data into the power generation prediction model to obtain the predicted surplus electric energy of the microgrid comprises the following steps:

[0042] Extract all weather data types from the historical weather data set;

[0043] Screening out invalid weather forecast data whose data type is different from all weather data types in the weather forecast data;

[0044] All the remaining weather forecast data are input into the power generation forecast model to obtain the predicted surplus electric energy of the microgrid.

[0045] Optionally, the energy storage server based on each of the shared energy storage devices constructs an energy scheduling blockchain network for the power distribution network, combines all the predicted energy storage data, and generates an energy scheduling plan for the power distribution network through the energy scheduling blockchain network, including the following steps:

[0046] The energy storage server of each shared energy storage device builds an energy dispatching blockchain network for the power distribution network;

[0047] Selecting a leader server among all the energy storage servers according to the server status of the energy storage servers;

[0048] Mark the shared energy storage device whose predicted energy storage data is lower than a preset electric energy data threshold as a target shared energy storage device, and mark the energy storage server of the target shared energy storage device as a target energy storage server;

[0049] Automatically generate an electric energy dispatch request for the target shared energy storage device through the target energy storage server, and upload the electric energy dispatch request to the energy dispatch blockchain network;

[0050] Using the leading server to broadcast the electric energy dispatch request to other energy storage servers other than the target energy storage server;

[0051] When the energy storage server receives the electric energy dispatch request, it generates electric energy dispatch information by combining the electric energy dispatch request and the predicted energy storage data stored locally, and uploads the electric energy dispatch information to the energy dispatch blockchain network;

[0052] All the electric energy dispatching information is aggregated through the leading server and an electric energy dispatching plan for the power distribution network is generated.

[0053] Optionally, the step of aggregating all the electric energy dispatching information and generating an electric energy dispatching plan for the power distribution network through the leading server comprises the following steps:

[0054] Aggregating all the electric energy dispatching information and generating an initial electric energy dispatching plan for the power distribution network through the leading server;

[0055] Broadcasting the initial power dispatching plan to all the energy storage servers through the leading server;

[0056] Using the consensus mechanism to verify whether the initial power dispatching plan has been verified;

[0057] If the initial electric energy dispatching scheme fails to be verified, the electric energy dispatching information regenerated by all the energy storage servers is obtained through the energy dispatching blockchain network, and the above scheme generation step and scheme verification step are repeated until the initial electric energy dispatching scheme is verified;

[0058] If the initial power dispatching plan passes the verification, the initial power dispatching plan will be used as the power dispatching plan for the power distribution network.

[0059] In a second aspect, the present invention also provides a power grid energy storage scheduling system based on deep learning and energy management, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the power grid energy storage scheduling method based on deep learning and energy management as described in the first aspect is implemented.

[0060] The beneficial effects of the present invention are:

[0061] By utilizing advanced machine learning algorithms such as deep convolutional neural networks and support vector machines, the present invention can more accurately predict the power load and the power generation of the microgrid. This high-precision prediction capability greatly improves the accuracy and reliability of power grid dispatching, and avoids the problem of power waste or shortage due to inaccurate predictions. In addition, the present invention fully considers multi-dimensional information such as historical power consumption data, real-time power consumption data, historical weather data, and weather forecast data. By comprehensively analyzing these data, it is possible to fully understand and predict the power demand and power generation capacity of the distribution area, thereby achieving more scientific and reasonable power dispatching. The present invention can also utilize the remaining stored power data of the shared energy storage device, which can effectively balance the supply and demand of the power grid, avoiding power waste or power grid instability caused by insufficient or excessive power generation of a single microgrid.

[0062] On the other hand, the present invention realizes data sharing and collaborative scheduling among various energy storage devices in the distribution network by constructing an energy scheduling blockchain network. The introduction of blockchain technology not only ensures the security and transparency of the data, but also improves the execution efficiency and reliability of the scheduling scheme. In this distributed scheduling mode, the energy storage device in each distribution area can flexibly adjust its own energy storage and discharge strategies according to the real-time power demand and power generation situation, so as to achieve the optimal scheduling of the entire power grid system. This flexible and intelligent scheduling method not only significantly improves the efficiency of power utilization and reduces energy waste, but also enhances the stability and risk resistance of the power grid system. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a flow chart of a grid energy storage scheduling method based on deep learning and energy management in one of the embodiments of the present application.

[0064] Figure 2 A schematic diagram of a process for predicting power consumption characteristics of power load in one embodiment of the present application.

[0065] Figure 3 It is a schematic diagram of a process of predicting surplus electric energy of a microgrid in one embodiment of the present application. DETAILED DESCRIPTION

[0066] The following will be combined with the drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the present application belong to the scope of protection of this application.

[0067] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0068] Reference Figure 1 , Figure 1 is a flow chart of a power grid energy storage scheduling method based on deep learning and energy management in one embodiment, Figure 1 The power grid energy storage scheduling method based on deep learning and energy management shown is applied to a distribution network, which includes multiple distribution areas, each of which contains multiple power loads and multiple microgrids built based on photovoltaic and wind power generation systems. Each distribution area is also equipped with a shared energy storage device, which is used to store the surplus electricity produced by all microgrids in the distribution area. The surplus electricity represents the part of the electricity produced by the microgrid in the same time period that exceeds the internal energy supply of the microgrid.

[0069] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1At least part of the steps in the above method may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps. Figure 1 As shown, a power grid energy storage scheduling method based on deep learning and energy management disclosed in the present invention specifically includes the following steps:

[0070] S101. For each power distribution area, obtain historical power consumption data and real-time power consumption data of all power loads in the power distribution area.

[0071] Among them, the historical power consumption data of all power loads are collected. Historical power consumption data usually includes information such as the power consumption, power consumption time, and power consumption mode of each power-consuming device. These data can be obtained from multiple channels such as smart meters, power consumption recording systems, and sensors of power-consuming devices. Smart meters are the most common data source. They can record the power consumption of each power-consuming device in real time and transmit the data to the central database. The power consumption recording system can provide more detailed historical power consumption data, including the trend of power consumption and peak power consumption hours. The sensors of power-consuming devices can monitor the operating status and power consumption of the equipment, providing more details for data collection. In addition to historical power consumption data, real-time power consumption data is also required. Real-time power consumption data refers to the power consumption of power-consuming devices at the current moment, including real-time power consumption, equipment operating status, etc. These data can be obtained in real time through smart meters and sensors, and transmitted to the data center through wireless networks or wired networks. In order to ensure the accuracy and completeness of the data, the collected data needs to be cleaned and preprocessed. The purpose of data cleaning is to remove missing values ​​and outliers to ensure the integrity and accuracy of the data. Missing values ​​can be processed by interpolation or filling, and outliers need to be identified and removed based on statistical methods or business rules. Data preprocessing also includes data normalization, which converts data of different dimensions to the same numerical range and eliminates the dimensional differences between different features.

[0072] S102. Based on a deep convolutional neural network and using historical electricity consumption data, a power consumption feature recognition model is constructed, and the real-time electricity consumption data is analyzed using the power consumption feature recognition model to predict the power consumption characteristics of the power load.

[0073] Among them, in the power consumption feature recognition, the historical power consumption data can be regarded as time series data, and the power consumption of each power-consuming device changes over time. The first step in building the power consumption feature recognition model is data preparation. The collected historical power consumption data is arranged in chronological order to form a time series data set. Then, the data set is divided into a training set and a test set. The training set is used to train the model, and the test set is used to evaluate the performance of the model. Next, the architecture of the deep convolutional neural network is designed. Convolutional neural networks usually include multiple convolutional layers, pooling layers, and fully connected layers. The convolutional layer is used to extract local features in the data, the pooling layer is used for dimensionality reduction and feature selection, and the fully connected layer is used for classification and regression. In power consumption feature recognition, a network architecture containing multiple convolutional layers and pooling layers can be designed to extract time series features and patterns in power consumption data. Then, the model is trained using the training set. During the training process, the model continuously adjusts parameters through the back propagation algorithm to minimize the prediction error. After the training is completed, the model is evaluated using the test set to verify the accuracy and robustness of the model. In this way, a power consumption feature recognition model can be obtained. Using this model, real-time electricity consumption data can be analyzed and the electricity consumption characteristics of the power load can be predicted.

[0074] S103. Calculate the predicted total power consumption of the power distribution area based on the predicted power consumption characteristics of all power loads.

[0075] Among them, first, the predicted power consumption characteristics of each power load are summarized to obtain the power consumption characteristic data set of the entire distribution area. This data set includes information such as the predicted power consumption and power consumption pattern of each power load in the future. Then, based on these power consumption characteristic data, the predicted total power consumption of the distribution area is calculated. Specifically, the predicted power consumption of each power load can be accumulated to obtain the total power consumption of the distribution area. In order to improve the accuracy of the prediction, a variety of methods can be used to correct and optimize the power consumption. For example, the predicted power consumption can be adjusted according to the trend and seasonal changes of historical data. In this way, a more accurate predicted total power consumption can be obtained.

[0076] S104. Obtain historical weather data and weather forecast data of the distribution area, as well as historical surplus power data of all microgrids.

[0077] Among them, historical weather data and historical surplus power data are in the same historical time period. Historical weather data usually includes meteorological elements such as temperature, humidity, wind speed, and rainfall. In addition to historical weather data, weather forecast data is also required. Weather forecast data is a weather forecast for a period of time in the future generated by numerical simulation and statistical methods based on meteorological models and observation data. These data can be obtained from the meteorological forecast system. In order to ensure the accuracy and consistency of the data, the weather forecast data needs to be preprocessed, including data cleaning, interpolation, normalization and other steps. In addition to weather data, historical surplus power data of all microgrids needs to be collected. Microgrid refers to a small power system composed of distributed energy, energy storage equipment, loads, etc. in a specific area. The historical surplus power data of microgrids includes information such as power generation, power consumption and surplus power of each microgrid in different time periods. These data can be obtained from channels such as the monitoring system and energy management system of the microgrid. In order to ensure the integrity and consistency of the data, the historical surplus power data needs to be cleaned and preprocessed, including removing missing values ​​and outliers, data normalization and other steps.

[0078] S105. Based on the support vector machine and using historical weather data and historical surplus power data, a power generation prediction model is constructed, the weather prediction data is preprocessed, and the preprocessed weather prediction data is input into the power generation prediction model to obtain the predicted surplus power of the microgrid.

[0079] Among them, in power generation prediction, historical weather data and historical surplus power data can be regarded as features and labels, and a power generation prediction model can be constructed through support vector machines. The first step in building a power generation prediction model is data preparation. The collected historical weather data and historical surplus power data are arranged in chronological order to form a time series data set. Then, the data set is divided into a training set and a test set. The training set is used to train the model, and the test set is used to evaluate the performance of the model. Next, the model structure of the support vector machine is designed. The support vector machine divides the data into different categories by finding an optimal hyperplane. In power generation prediction, historical weather data can be used as features and historical surplus power data as labels. The relationship between features and labels is learned by support vector machines to build a power generation prediction model. Then, the model is trained using the training set. During the training process, the model continuously adjusts parameters through the optimization algorithm to minimize the prediction error. After the training is completed, the model is evaluated using the test set to verify the accuracy and robustness of the model. In this way, a power generation prediction model can be obtained. Using this model, weather forecast data can be analyzed to predict the surplus power of the microgrid.

[0080] S106. Based on the predicted total electricity consumption, the predicted surplus electric energy of all microgrids and the remaining stored electric energy of the shared energy storage device, the predicted energy storage data of the shared energy storage device is calculated and stored in the energy storage server carried by the shared energy storage device.

[0081] The predicted energy storage data of the shared energy storage device can be simply calculated by subtracting the predicted total electricity consumption from the sum of the predicted surplus energy and the remaining stored energy. After the calculation is completed, the predicted energy storage data is stored in the energy storage server carried by the shared energy storage device. The energy storage server is a cloud server dedicated to storing and managing energy storage data.

[0082] S107. Build an energy dispatching blockchain network for the distribution network based on the energy storage server of each shared energy storage device, combine all predicted energy storage data and generate an energy dispatching plan for the distribution network through the energy dispatching blockchain network.

[0083] Among them, blockchain is a decentralized distributed ledger technology that ensures the security and immutability of data through encryption algorithms and consensus mechanisms. In energy scheduling, blockchain technology can be used to build a decentralized energy scheduling network, and the energy storage server of each shared energy storage device is used as a node of the blockchain network. Each node exchanges data and reaches consensus through the blockchain protocol to ensure the security and consistency of data. The first step in building an energy scheduling blockchain network is node deployment. The energy storage server of each shared energy storage device is deployed as a blockchain node, and the nodes use encrypted communication protocols for data transmission and consensus.

[0084] Next, design the consensus mechanism and smart contract of the blockchain network. The consensus mechanism is used to ensure the consistency and security of data in the blockchain network. Common consensus mechanisms include proof of work, proof of equity, etc. Smart contracts are automated programs running on the blockchain that are used to execute and verify transactions. In energy scheduling, smart contracts can be designed to automatically generate and execute power scheduling plans. Then, all predicted energy storage data are uploaded to the blockchain network. Each node packages its own predicted energy storage data into blocks and verifies and records them through the consensus mechanism. In this way, the security and immutability of the data can be ensured. Next, the power scheduling plan for the distribution network is generated through smart contracts. Smart contracts automatically calculate and optimize the power scheduling plan based on the predicted energy storage data and energy demand. Specifically, the distribution and scheduling of electricity can be reasonably arranged according to electricity demand, power generation and storage capacity to ensure stable and efficient power supply in the distribution area.

[0085] In one embodiment, referring to Figure 2Based on the deep convolutional neural network and using the historical power consumption data to build a power consumption feature recognition model, the power consumption feature recognition model is used to analyze the real-time power consumption data, and the predicted power consumption features of the power load are predicted, which includes the following steps:

[0086] S201. Construct an initial electricity consumption feature recognition model based on a deep convolutional neural network.

[0087] S202. Preprocess all historical electricity consumption data into a historical electricity consumption data set.

[0088] Among them, all historical electricity consumption data are pre-marked with corresponding historical electricity consumption characteristics.

[0089] S203. Perform feature correlation analysis on the historical electricity consumption features of the historical electricity consumption data set to obtain an electricity consumption feature classification matrix of the historical electricity consumption data set.

[0090] S204. Use the electricity consumption feature classification matrix to train the initial electricity consumption feature recognition model to obtain a trained electricity consumption feature recognition model.

[0091] S205. Analyze the real-time power consumption data using the power consumption feature recognition model to predict the power consumption features of the power load.

[0092] In this embodiment, a deep convolutional neural network (DCNN) is a powerful tool for time series data analysis, and its multi-layer structure can effectively extract and learn complex features in the data. In the recognition of electricity consumption features, the advantage of DCNN is that it can extract feature information from electricity consumption data layer by layer through hierarchical convolution operations, thereby achieving accurate classification and recognition of electricity consumption features. First, it is necessary to define the hierarchical structure of the network, including an input layer, multiple convolutional layers, a pooling layer, a fully connected layer, and an output layer. The input layer is responsible for receiving electricity consumption data, which may include time series data such as voltage, current, and power. The convolution layer extracts local features through convolution operations, and the size and number of convolution kernels are hyperparameters that need to be adjusted. The pooling layer reduces the data dimension by downsampling while retaining important features. Commonly used pooling operations include maximum pooling and average pooling. The fully connected layer synthesizes the extracted features, and the output layer gives the final electricity consumption feature recognition results.

[0093] In order to improve the performance of the model, a batch normalization layer can be added after the convolution layer, which helps to speed up the training process and stabilize the training effect of the model. Activation functions, such as ReLU (rectified linear unit), are used to enhance the nonlinear expression ability of the model, so that the model can fit complex power consumption feature patterns. The training process of the model uses large-scale historical power consumption data, and adjusts the network parameters through the back propagation algorithm so that the model can accurately identify and classify power consumption features. The back propagation algorithm guides the update direction and amplitude of the parameters by calculating the gradient of the loss function. Commonly used optimization algorithms include stochastic gradient descent (SGD) and its variants such as the Adam optimization algorithm. During the training process, hyperparameters such as learning rate, batch size, number of network layers, etc. need to be continuously adjusted to find the optimal model configuration. In addition, in order to avoid overfitting, regularization techniques such as L2 regularization or Dropout can be used to randomly discard the output of some neurons to enhance the generalization ability of the model. After multiple iterations of training and verification, a high-performance power consumption feature recognition model is finally obtained, which can accurately identify and classify features in power consumption data.

[0094] After building the model, all historical electricity consumption data need to be preprocessed into a historical electricity consumption data set. The purpose of preprocessing is to clean the data, eliminate noise, and standardize the data to improve the effect and accuracy of model training. First, collect all historical electricity consumption data, which usually includes information such as voltage, current, power, and timestamp. Next, clean the data to remove missing values ​​and outliers. Missing values ​​can be processed by interpolation or filling, and outliers need to be identified and eliminated based on statistical methods or business rules. Then, normalize the data to convert data of different dimensions to the same numerical range, such as standardizing the data to between 0 and 1. This step helps to eliminate the dimensional differences between different features and prevent certain features from having too much impact on the model during training.

[0095] Each piece of historical electricity consumption data is pre-labeled with the corresponding historical electricity consumption features. These features can be the load type of the power load, the historical electricity consumption pattern, etc. The labeling process can be completed by an automatic labeling tool to ensure that each piece of data has a clear feature label. In order to further improve the quality and consistency of the data, data enhancement technology can be used to generate more training samples by performing random transformations on the original data, such as translation, rotation, and scaling. Data enhancement can not only increase the amount of data, but also improve the robustness of the model, so that it can better adapt to different electricity consumption scenarios and feature changes. Finally, the preprocessed historical electricity consumption data set will be used as the basic data for model training. These data not only contain rich information on electricity consumption characteristics, but also have been strictly cleaned and labeled to ensure high quality and high credibility of the data.

[0096] The purpose of feature correlation analysis is to identify and quantify the relationship between different electricity consumption features, so as to construct an electricity consumption feature classification matrix. First, it is necessary to perform statistical analysis on each feature in the data set and calculate its basic statistics. Then, correlation analysis methods such as Pearson correlation coefficient and Spearman correlation coefficient are used to calculate the correlation between each feature. Correlation analysis can help identify those features that have a greater impact on electricity consumption patterns and exclude redundant or irrelevant features. Then, using dimensionality reduction techniques such as principal component analysis (PCA) or factor analysis (FA), the high-dimensional feature data is reduced to a low-dimensional space to facilitate subsequent classification and model training. Through these analysis methods, an electricity consumption feature classification matrix can be constructed, which reflects the correlation and importance between different electricity consumption features. The construction of the feature classification matrix helps to improve the training efficiency and accuracy of the model, avoid the interference of redundant features during the training process, and enhance the generalization ability of the model.

[0097] Using the electricity consumption feature classification matrix to train the initial electricity consumption feature recognition model is a key step. Specifically, the preprocessed historical electricity consumption data set and the electricity consumption feature classification matrix can be used as the input data of the model. The data set is divided into a training set and a validation set in a certain proportion. The training set is used to adjust and optimize the model parameters, and the validation set is used to evaluate the performance of the model. During the model training process, the cross entropy loss function or the mean square error loss function is used to measure the difference between the model's prediction results and the actual labels. The gradient of the loss function to the model parameters is calculated through the back propagation algorithm, and the model parameters are updated using the gradient descent algorithm or its variants, such as the Adam optimization algorithm. In order to avoid overfitting, regularization techniques such as L2 regularization or Dropout can be used to randomly discard the output of some neurons to enhance the generalization ability of the model. During the training process, hyperparameters such as learning rate, batch size, number of network layers, etc. need to be continuously adjusted to find the optimal model configuration.

[0098] Through multiple iterations of training and validation, we finally get a high-performance electricity feature recognition model that can accurately identify and classify features in electricity data. During the training process, we can use the cross-validation method to divide the data set into multiple subsets, use one of the subsets as the validation set in turn, and the remaining subsets as the training set to ensure the stability and generalization ability of the model under different data distributions. In addition, in order to further improve the performance of the model, we can use the ensemble learning method to train multiple different models and fuse their prediction results to obtain more accurate and robust prediction results. Ensemble learning can be implemented in many ways, such as voting method, weighted average method, etc. Different weights are given according to the performance of different models, and the advantages of each model are combined to improve the accuracy and stability of the final prediction results.

[0099] The trained power consumption feature recognition model can be used to analyze and predict real-time power consumption data. First, the real-time collected power consumption data is preprocessed in a similar way to the preprocessing of historical data, including data cleaning and normalization. Then, the preprocessed real-time power consumption data is input into the power consumption feature recognition model. The model processes the input data layer by layer through the forward propagation process, extracts features, and performs classification or regression to output the prediction results. The prediction results can be feature information such as the type of power load and the power consumption mode. In order to verify the accuracy of the prediction results, the prediction results can be compared with the actual power consumption data to calculate error indicators such as mean square error and absolute error. If the error is within an acceptable range, it means that the prediction performance of the model is good. Otherwise, it is necessary to further optimize the model parameters or adjust the model structure.

[0100] In one implementation manner, performing feature correlation analysis on historical power consumption features of a historical power consumption data set to obtain a power consumption feature classification matrix of the historical power consumption data set includes the following steps:

[0101] Extract historical electricity consumption features from historical electricity consumption datasets;

[0102] The connection function is used to calculate the power consumption feature correlation of the historical power consumption features of different feature types;

[0103] The decision tree learning algorithm based on information entropy is used to analyze the correlation of electricity consumption features, and the electricity consumption feature classification matrix of the historical electricity consumption data set is obtained.

[0104] In this embodiment, after extracting the historical electricity consumption features, the next step is to calculate the correlation between the electricity consumption features of each different feature type. The calculation of the correlation helps to understand the relationship between different features and provide a reference for subsequent feature selection and model optimization. Using a connection function is a common method for calculating the correlation. First, select an appropriate connection function. Common connection functions include the Pearson correlation coefficient, the Spearman rank correlation coefficient, and mutual information. The Pearson correlation coefficient is used to measure the linear relationship between two variables, and the value range is [-1,1], where 1 represents a perfect positive correlation, -1 represents a perfect negative correlation, and 0 represents no correlation. The Spearman rank correlation coefficient is used to measure the monotonic relationship between two variables and is applicable to nonlinear relationships. Mutual information is used to measure the dependency between two variables and is not limited by linear or nonlinear relationships.

[0105] In the specific implementation process, all eigenvectors need to be standardized first to eliminate the influence between different eigendimensions. The standardization method can choose mean-variance standardization or minimum-maximum standardization. Mean-variance standardization subtracts the mean from the eigenvalue and divides it by the standard deviation, so that the mean of the standardized eigenvalue is 0 and the standard deviation is 1. Minimum-maximum standardization scales the eigenvalue to the range of [0,1]. Taking the Spearman rank correlation coefficient as an example, the eigenvalues ​​need to be converted to ranks first, and then the correlation coefficient between the ranks is calculated. The calculation formula is:

[0106] ,

[0107] Where di represents the rank difference between two historical electricity consumption features, and n represents the number of historical electricity consumption features.

[0108] After calculating the correlation of electricity consumption features, the next step is to use the decision tree learning algorithm based on information entropy to analyze these correlations and obtain the electricity consumption feature classification matrix of the historical electricity consumption data set. The decision tree is a commonly used classification and regression algorithm that recursively divides the data set into subsets to construct a tree structure. Information entropy is used to measure the uncertainty of the data set, and the decision tree selects the best split point by maximizing information gain. All historical electricity consumption features are constructed into an initial decision tree based on the correlation of electricity consumption features. Each node of the decision tree represents a feature, and each edge represents a split condition for a feature value. The root node represents the entire data set, and the leaf node represents the classification result. In the process of building a decision tree, the information entropy of each feature is first calculated. The calculation formula for information entropy is:

[0109] ,

[0110] Where D represents the historical electricity consumption dataset, and pi represents the probability of the i-th type of historical electricity consumption feature. Information gain represents the reduction in information entropy after feature segmentation, and the calculation formula is:

[0111] ,

[0112] Where A represents the historical electricity consumption characteristics, and Dv represents the subset of the historical electricity consumption characteristics A with the value v.

[0113] In the specific implementation process, the information gain of each feature is first calculated, and the feature with the largest information gain is selected as the split node. Then, the data set is divided into several subsets according to the value of the feature, and a sub-decision tree is recursively constructed for each subset until all features are used or the samples in the subset belong to the same class. After the decision tree is built, the decision tree can be pruned according to the feature association. The purpose of pruning is to remove redundant nodes and improve the generalization ability of the model. Common pruning methods include pre-pruning and post-pruning. Pre-pruning is to limit the growth of the tree by setting parameters such as the maximum depth and the minimum number of samples during the construction of the decision tree; post-pruning is to remove unnecessary nodes by cross-validation and other methods after the decision tree is built. Finally, the decision tree is used to analyze the association of electricity consumption features to obtain an electricity consumption feature classification matrix. Each row of the classification matrix represents a feature, each column represents a classification result, and the value in the matrix represents the distribution of the feature under different classification results. Through this classification matrix, the contribution of different features in the classification of electricity consumption features can be intuitively understood, which provides a basis for subsequent feature selection and model optimization.

[0114] In one embodiment, using the power consumption feature classification matrix to train the initial power consumption feature recognition model to obtain the trained power consumption feature recognition model includes the following steps:

[0115] The particle swarm optimization algorithm is used to select the optimal power consumption features from the power consumption feature classification matrix;

[0116] Based on the optimal electricity consumption characteristics, the optimal feature matrix is ​​constructed by weighted least square method;

[0117] The optimal feature matrix is ​​input into the initial power consumption feature recognition model, and the bias item parameters of all fully connected layers in the initial power consumption feature recognition model are calculated;

[0118] Calculate the parameter error of the bias item parameters through the preset optimal parameters;

[0119] If the parameter error is greater than or equal to the preset error threshold, all the above steps are executed again until the parameter error is less than the error threshold;

[0120] If the parameter error is less than the error threshold, the model training process of the initial power consumption feature recognition model is completed, and a trained power consumption feature recognition model is obtained.

[0121] In this embodiment, the particle swarm optimization algorithm is first used to select the optimal power consumption features from the power consumption feature classification matrix. After determining the optimal power consumption features, it is necessary to construct the optimal feature matrix through the weighted least squares method (Weighted Least Squares, WLS). The weighted least squares method is a linear regression method that minimizes the sum of squared errors by assigning different weights to different optimal power consumption features, thereby improving the fitting accuracy of the model. In specific implementation, it is first necessary to construct a design matrix X, in which each row represents a sample and each column represents a feature. Then, a weight matrix W is constructed, which is a diagonal matrix in which the elements on the diagonal represent the weight of each sample. Through the weighted least squares method, the optimal feature matrix can be obtained. , and its calculation formula is:

[0122] ,

[0123] The optimal feature matrix constructed by this method can maximize the reflection of the relationship between electricity consumption characteristics and electricity consumption, thus providing a solid foundation for subsequent model training.

[0124] The optimal feature matrix is ​​input into the initial power consumption feature recognition model. Through forward propagation, the optimal feature matrix is ​​weighted summed and nonlinearly transformed at each layer, and finally the prediction result is obtained through the output layer. In order to calculate the bias parameters of each layer, back propagation and gradient descent are required. Back propagation updates the weights and bias items by calculating the partial derivatives of the loss function (such as mean square error) with respect to the parameters of each layer. The specific formula is:

[0125] ,

[0126] Where: is the bias term after the t+1th iteration, is the bias term after the tth iteration, is the learning rate and L is the loss function.

[0127] After calculating the bias item parameters of all connection layers in the initial power consumption feature recognition model, it is necessary to calculate the parameter errors of these bias item parameters through the preset optimal parameters. Parameter error refers to the difference between the actual calculated bias item parameters and the expected optimal parameters. This difference can be measured in a variety of ways, the most common of which is to use the mean squared error (MSE) or absolute error (Absolute Error). In specific implementation, it is necessary to first define the preset optimal parameters. These parameters are usually obtained through prior knowledge or empirical data, and represent the parameter values ​​that the model should have under ideal conditions. Then, the error between the actual bias item parameters and the preset optimal parameters is calculated. If the error is greater than or equal to the preset error threshold, all the above steps need to be re-executed, including reconstructing the optimal feature matrix, inputting the initial power consumption feature recognition model, calculating the bias item parameters of the connection layer, etc., until the errors of all parameters are less than the error threshold.

[0128] Through this iterative process, the model parameters can be continuously optimized, the parameter errors can be gradually reduced, and the expected accuracy requirements can be finally achieved. If the errors of all parameters are less than the error threshold, it means that the model training process is completed and a trained power consumption feature recognition model is obtained.

[0129] In one embodiment, selecting the optimal power consumption feature from the power consumption feature classification matrix using a particle swarm optimization algorithm includes the following steps:

[0130] Randomly select initial electricity consumption features from the electricity consumption feature classification matrix;

[0131] Initializing the initial power consumption characteristics into two-dimensional feature data using preset particle swarm algorithm parameters, and constructing an initial particle swarm based on the two-dimensional feature data;

[0132] The initial particle swarm data is optimized multiple times through the particle swarm optimization algorithm until the initial particle swarm is optimized to the optimal particle swarm;

[0133] All initial power consumption features in the optimal particle swarm are taken as the optimal power consumption features.

[0134] In this embodiment, after obtaining the power consumption feature classification matrix, the initial power consumption feature is randomly selected from the power consumption feature classification matrix, and the particle swarm optimization (PSO) algorithm is used to select the optimal power consumption feature. PSO is an optimization algorithm that simulates swarm intelligence and finds the optimal solution by simulating the foraging behavior of bird flocks. Each particle represents a candidate solution, and the solution space is searched by continuously updating the position and speed. Specifically, the initial power consumption feature is first initialized to two-dimensional feature data using the preset particle swarm algorithm parameters, and an initial particle swarm is constructed based on the two-dimensional feature data. Each particle in the initial particle swarm represents a feature subset, and the dimension of the particle is equal to the number of two-dimensional feature data. The initial position and speed of the particle can be randomly generated or initialized according to the feature importance of the feature classification matrix. Each particle has a fitness value, which is used to measure the pros and cons of the feature subset. The fitness value can be calculated based on indicators such as classification accuracy and information gain.

[0135] In the specific implementation process, first determine the size of the particle swarm and the number of iterations. The size of the particle swarm determines the coverage of the search space, and the number of iterations determines the termination condition of the algorithm. Generally speaking, a larger particle swarm and more iterations can improve the search accuracy, but also increase the computational complexity. Next, update the particle speed and position. The particle speed update formula is:

[0136] ,

[0137] Where: vi(t) represents the velocity of particle i in the tth generation, xi(t) represents the position of particle i in the tth generation, pi represents the historical best position of particle i, pg represents the global best position, w represents the inertia weight, c1 and c2 represent acceleration constants, and r1 and r2 are random numbers between [0, 1].

[0138] The particle position update formula is:

[0139] ,

[0140] Through the above formula, each particle moves continuously in the search space and gradually approaches the optimal solution. In each iteration, the fitness value of each particle is updated and the global optimal position is recorded. When the maximum number of iterations is reached or the fitness value is no longer significantly improved, the algorithm terminates and returns the global optimal position as the optimal feature subset. Through the PSO algorithm, the optimal power consumption features are selected from the power consumption feature classification matrix, which can effectively improve the accuracy and efficiency of feature selection and provide a high-quality feature set for subsequent model training and prediction.

[0141] In one embodiment, referring to Figure 3Based on the support vector machine and using historical weather data and historical surplus power data to build a power generation prediction model, the weather prediction data is preprocessed, and the preprocessed weather prediction data is input into the power generation prediction model to obtain the predicted surplus power of the microgrid, which includes the following steps:

[0142] S301. Construct an initial power generation prediction model based on support vector machine.

[0143] S302. Calculate the data correlation between different types of historical weather data and historical surplus electric energy data using a grey correlation algorithm.

[0144] S303. Integrate all historical weather data with a data correlation greater than a preset correlation threshold into a historical weather data set.

[0145] S304. Iteratively train the initial power generation prediction model in combination with the historical weather data set and the historical surplus power data until the model parameters in the initial power generation prediction model are trained to the optimal model parameters, thereby obtaining a trained power generation prediction model.

[0146] S305 pre-processes the weather forecast data, and inputs the pre-processed weather forecast data into a power generation prediction model to obtain the predicted surplus electric energy of the microgrid.

[0147] In this embodiment, the support vector machine is a supervised learning model widely used in classification and regression analysis. The core idea is to find an optimal hyperplane in a high-dimensional space to segment the data, thereby realizing the prediction of new data. After constructing the initial power generation prediction model, the relationship between historical weather data and historical surplus power data needs to be analyzed next. The gray correlation algorithm is an effective analysis tool that can quantify the degree of association between different variables. The historical weather data can be paired with the corresponding historical surplus power data to form a multidimensional data set. The gray correlation algorithm measures the degree of association between the variables by calculating the gray correlation between them. Specifically, the gray correlation is determined by calculating the relative degree of change between the variables. First, the data needs to be dimensionless so that data of different dimensions can be compared. Then, the difference sequence between each variable and the reference variable (historical surplus power data) is calculated, and the gray correlation is obtained by solving the absolute value and the maximum and minimum values ​​of the difference sequence. The calculated gray correlation can reflect the degree of association between each weather variable and the historical surplus power data. The higher the correlation, the greater the impact of the weather variable on the surplus power. Using this method, key weather factors such as temperature, humidity, wind speed, etc. can be screened out. These key weather factors will directly affect the power generation efficiency of photovoltaic and wind power generation, and therefore have a greater impact on surplus electricity.

[0148] After calculating the grey correlation between each weather variable and the historical surplus power data, the next step is to filter out weather data with a high correlation. Set a correlation threshold, and only weather variables with a correlation greater than the threshold will be included in the historical weather data set. The setting of this threshold can be adjusted according to actual conditions and experience, and usually a value that can significantly distinguish between high correlation and low correlation is selected. The filtered high-correlation weather variables are integrated into a new historical weather data set. This data set contains key weather factors that have a greater impact on surplus power. In this way, data redundancy can be reduced, and data quality and model training efficiency can be improved.

[0149] After screening out the historical weather data with high correlation, these data are combined with the historical surplus power data to perform iterative training of the initial power generation forecast model. First, the integrated historical weather data set is paired with the historical surplus power data to form a new training data set. This data set will be used to iteratively train the initial power generation forecast model to optimize the model parameters. The iterative training process is a process of continuously adjusting and optimizing model parameters. First, the new training data set is input into the initial power generation forecast model and trained using the support vector machine algorithm. During the training process, the model will continuously adjust the parameters based on the input data to minimize the prediction error. This process usually requires multiple iterations, and each iteration adjusts the model parameters and calculates a new prediction error.

[0150] In order to improve the training efficiency and prediction accuracy of the model, a cross-validation method can be used. The training data set is divided into multiple subsets, and one of the subsets is used as the validation set each time, and the remaining subsets are used as the training set. In this way, the performance of the model can be evaluated more comprehensively to avoid overfitting and underfitting problems. After each iterative training, the prediction performance of the model is evaluated, and the prediction error and evaluation index are calculated. By analyzing these indicators, the degree of optimization of the model parameters can be judged. If the prediction performance of the model reaches the expected goal, the iterative training is stopped to obtain the final power generation prediction model. Otherwise, the model parameters are continued to be adjusted and the next round of iterative training is carried out until the model performance reaches the optimal level. After obtaining the trained power generation prediction model, the model can be used for actual prediction.

[0151] In one embodiment, preprocessing the weather forecast data and inputting the preprocessed weather forecast data into a power generation prediction model to obtain the predicted surplus power of the microgrid includes the following steps:

[0152] Extract all weather data types from the historical weather dataset;

[0153] Filter out invalid weather forecast data whose data types are different from all weather data types in the weather forecast data;

[0154] All remaining weather forecast data are input into the power generation forecast model to obtain the predicted surplus power of the microgrid.

[0155] In this embodiment, the collated historical weather data set is classified and extracted using a data analysis tool or a programming language. Specifically, a script can be written to traverse each record in the historical weather data set to identify and extract the weather data type therein. After extracting the historical weather data type, it is necessary to screen out the invalid weather forecast data in the weather forecast data whose data type is different from all weather data types. The purpose of this step is to remove invalid data that does not meet the requirements and ensure that the data input into the power generation prediction model is consistent and accurate. After screening out the invalid weather forecast data, the remaining data will be retained as valid weather forecast data. These valid data will be used for subsequent power generation prediction model input. In this way, the predicted surplus electric energy of the microgrid can be predicted.

[0156] In one embodiment, an energy dispatching blockchain network for a power distribution network is constructed based on an energy storage server of each shared energy storage device, and an energy dispatching scheme for the power distribution network is generated through the energy dispatching blockchain network by combining all predicted energy storage data, including the following steps:

[0157] Building an energy dispatching blockchain network for power distribution networking based on the energy storage server of each shared energy storage device;

[0158] Select a leader server among all energy storage servers based on the server status of the energy storage servers;

[0159] Marking a shared energy storage device whose predicted energy storage data is 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;

[0160] Automatically generate an energy dispatch request for the target shared energy storage device through the target energy storage server, and upload the energy dispatch request to the energy dispatch blockchain network;

[0161] Using the leader server to broadcast the power dispatch request to other energy storage servers other than the target energy storage server;

[0162] When the energy storage server receives the power dispatch request, it generates power dispatch information by combining the power dispatch request and the locally stored predicted energy storage data, and uploads the power dispatch information to the energy dispatch blockchain network;

[0163] All power dispatch information is aggregated through the leading server and a power dispatch plan for the distribution network is generated.

[0164] In this embodiment, an energy dispatching blockchain network for distribution network is constructed based on the energy storage server of each shared energy storage device. The core of this step is to use the distributed ledger and smart contract functions of blockchain technology to achieve efficient management and dispatch of shared energy storage devices. When implementing this step, it is first necessary to configure an energy storage server for each shared energy storage device, which is responsible for managing and monitoring the status and data of each energy storage device. Next, these energy storage servers are connected using blockchain technology to build a decentralized energy dispatching network. The distributed ledger characteristics of blockchain technology can ensure data sharing and synchronization between all servers, while its immutability can ensure the security and credibility of data. In specific implementation, a suitable blockchain platform, such as Ethereum or Hyperledger Fabric, can be selected to deploy a private chain or alliance chain. Each energy storage server will be added to the blockchain network as a node, and the consistency of the ledger will be maintained between nodes through a consensus mechanism (such as PoW, PoS or PBFT). Smart contracts are used to automatically execute business logic related to energy dispatching, such as the generation and processing of power dispatching requests.

[0165] After building the energy dispatch blockchain network, the next step is to elect the leader server among all energy storage servers based on the server status of the energy storage servers. The leader server plays a role of coordination and management in the entire network, responsible for processing and distributing energy dispatch requests to ensure the efficient operation of the system. When implementing this step, it is first necessary to define the evaluation criteria for server status, which may include the server's computing power, network latency, storage capacity, online time, etc. By comprehensively evaluating these criteria, the server with the best performance can be selected as the leader server. In specific implementation, a decentralized election algorithm, such as Raft or Paxos, can be used, which can efficiently elect leaders in distributed systems.

[0166] During the election process, each energy storage server will conduct a self-evaluation based on its own status information and broadcast the evaluation results to other servers. After receiving this information, all servers will vote according to the preset election rules and finally select the server with the most votes as the leader server. All operations and results during the election process will be recorded on the blockchain to ensure transparency and traceability. After the leader server is elected, this server will be responsible for coordinating and managing the entire energy dispatch network, processing power dispatch requests and distributing them to other servers to ensure efficient operation of the system and reasonable distribution of power.

[0167] After the leader server is elected, the next step is to identify the shared energy storage devices that require power scheduling based on the predicted energy storage data. Specifically, the shared energy storage devices whose predicted energy storage data is lower than the 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. When implementing this step, you first need to obtain the predicted energy storage data of each shared energy storage device, which is usually predicted by the energy storage server based on historical data and current status. Next, set an energy data threshold, which can be set according to the needs and safety requirements of the system. For example, a minimum energy storage level can be set, and when the predicted energy storage data is lower than this level, the device needs to be scheduled for power.

[0168] After marking the target shared energy storage device and the target energy storage server, the next step is to automatically generate the power dispatch request of the target shared energy storage device through the target energy storage server and upload these requests to the energy dispatch blockchain network. The core of this step is to use smart contracts and automation technology to realize the automatic generation and processing of power dispatch requests. When implementing this step, it is first necessary to deploy smart contracts on the target energy storage server. These contracts are used to define the generation rules and processes of power dispatch requests. Smart contracts can automatically generate power dispatch requests based on the predicted energy storage data and current status of the target shared energy storage device. For example, when the predicted energy storage data is lower than the set threshold, the smart contract will trigger the generation of power dispatch requests, specifying information such as the amount of electricity and time to be dispatched. The generated power dispatch requests will be recorded on the blockchain to ensure the security and immutability of the data. Through the blockchain network, the power dispatch requests will be broadcast to all relevant nodes to ensure that all energy storage servers can receive and process these requests.

[0169] After generating and uploading the power dispatch request, the next step is to use the leader server to broadcast these requests to other energy storage servers other than the target energy storage server. When implementing this step, the leader server will regularly scan the power dispatch requests on the blockchain and identify the requests that need to be broadcast. Next, the leader server broadcasts these requests to other energy storage servers through the blockchain network. During the broadcast process, the leader server will select the best broadcast path based on the network topology and node status to ensure that the request can be quickly transmitted. In specific implementation, the P2P communication protocol of the blockchain network can be used to send power dispatch requests to other servers through the message passing mechanism. After receiving the request, each server will process it according to the content of the request and its own status, and feedback the processing results to the leader server.

[0170] After the energy storage server receives the power dispatch request, the next step is to combine the request content with the locally stored predicted energy storage data to generate power dispatch information and upload this information to the energy dispatch blockchain network. The core of this step is to use local data and request content to generate accurate power dispatch information to ensure the efficient operation of the system. When implementing this step, the energy storage server will first parse the received power dispatch request and extract the dispatch information, such as the amount of power and time to be dispatched. Next, the server compares and analyzes this information with the locally stored predicted energy storage data to generate specific power dispatch information. This process can be achieved by writing data processing scripts or calling smart contracts. The generated power dispatch information will be recorded on the blockchain to ensure the security and immutability of the data. Through the blockchain network, the power dispatch information will be broadcast to all relevant nodes to ensure that all energy storage servers can receive and process this information.

[0171] After all energy storage servers generate and upload power dispatch information, the last step is to aggregate this information through the leader server and generate a power dispatch plan for the distribution network. The core of this step is to use the coordination and management functions of the leader server to integrate all dispatch information and generate a global power dispatch plan to ensure the efficient operation of the system and the reasonable distribution of electricity. When implementing this step, the leader server will regularly scan the power dispatch information on the blockchain and collect all relevant dispatch data. Next, the leader server will aggregate and analyze this data to generate a global power dispatch plan. This process can be achieved by calling smart contracts or writing data processing scripts. The generated power dispatch plan will be recorded on the blockchain to ensure the security and immutability of the data. Through the blockchain network, the power dispatch plan will be broadcast to all relevant nodes to ensure that all energy storage servers can receive and execute these plans.

[0172] In one embodiment, the leading server aggregates all power dispatch information and generates a power dispatch plan for the power distribution network, including the following steps:

[0173] The leading server aggregates all power dispatch information and generates an initial power dispatch plan for the power distribution network;

[0174] Broadcast the initial power dispatch plan to all energy storage servers through the leader server;

[0175] Use the consensus mechanism to verify whether the initial power dispatch plan has been verified;

[0176] If the initial power dispatching scheme fails to be verified, the power dispatching information regenerated by all energy storage servers is obtained through the energy dispatching blockchain network, and the above scheme generation steps and scheme verification steps are repeated until the initial power dispatching scheme is verified;

[0177] If the initial power dispatching plan is verified, the initial power dispatching plan will be used as the power dispatching plan for the distribution network.

[0178] In this embodiment, when the leading server aggregates all the power dispatching information and generates the initial power dispatching plan for the distribution network, the leading server will first collect the power dispatching information uploaded by all energy storage servers from the blockchain network. This information includes the current energy storage status of each shared energy storage device, predicted energy storage data, and previously generated power dispatching requests. The leading server will conduct a comprehensive analysis of this data, taking into account the energy storage requirements, power supply capacity, and overall power balance of each energy storage device. By calling smart contracts or preset algorithms, the leading server integrates this information to generate a preliminary power dispatching plan. This plan aims to optimize the distribution of electricity and ensure that each energy storage device can achieve reasonable scheduling and use of electricity while meeting the minimum energy storage requirements.

[0179] After the initial power dispatch plan is generated, the leading server broadcasts the plan to all energy storage servers. The broadcast process uses the P2P communication protocol of the blockchain network to ensure that each energy storage server can receive the initial power dispatch plan in a timely manner. After receiving the plan, the energy storage server will verify it locally to check whether the plan meets its own energy storage status and needs. Each step in the broadcast process is recorded on the blockchain to ensure data transparency and security. In this way, the initial power dispatch plan can be quickly transmitted to every node in the entire network, ensuring that all energy storage servers can participate in the verification and execution of the plan.

[0180] Using the consensus mechanism to verify whether the initial power dispatch plan is passed is a key step. Consensus mechanisms such as PoW, PoS or PBFT ensure that all energy storage servers reach a consensus on the initial power dispatch plan. After receiving the plan, each energy storage server will verify it according to its own energy storage status and needs, and feedback the verification results to the blockchain network. The consensus mechanism will summarize and analyze all feedback results according to the preset rules, and finally determine whether the initial power dispatch plan is verified. This process ensures the rationality and executability of the plan, and at the same time uses the distributed ledger characteristics of the blockchain to ensure the transparency and credibility of the verification process.

[0181] If the initial power dispatch scheme fails to pass the verification, the power dispatch information regenerated by all energy storage servers is obtained through the energy dispatch blockchain network, and the above scheme generation steps and scheme verification steps are repeated until the initial power dispatch scheme is verified. In this case, the energy storage server will regenerate the power dispatch information according to the new energy storage status and demand, and upload it to the blockchain network. The leading server summarizes this information again, generates a new initial power dispatch scheme, and broadcasts it to all energy storage servers for verification. This cycle continues until all energy storage servers reach a consensus on the scheme, ensuring that the final generated power dispatch scheme can meet the overall needs of the system and the specific requirements of each energy storage device.

[0182] If the initial power dispatching scheme is verified, the initial power dispatching scheme will be used as the power dispatching scheme for the distribution network. In this case, the verified initial power dispatching scheme is regarded as the final power dispatching scheme and is recorded and published in the blockchain network. All energy storage servers will perform power dispatching operations according to the scheme to ensure the reasonable distribution and use of electric energy. In this way, the entire distribution network system can achieve efficient power management and dispatching, ensure the stable operation of each shared energy storage device and the reasonable distribution of electric energy. At the same time, the distributed ledger and smart contract functions of blockchain technology are used to ensure the transparency and traceability of the entire process, providing strong support for system optimization and management.

[0183] The present invention also discloses a power grid energy storage scheduling system based on deep learning and energy management, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the power grid energy storage scheduling method based on deep learning and energy management as described in any one of the above embodiments.

[0184] Among them, the processor can adopt a central processing unit (CPU). Of course, according to actual usage, other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be adopted. The general-purpose processor can adopt a microprocessor or any conventional processor, etc., and this application does not impose any restrictions on this.

[0185] Among them, the memory can be an internal storage unit of a computer device, such as a hard disk or memory of a computer device, or an external storage device of a computer device, such as a plug-in hard disk, a smart memory card (SMC), a secure digital card (SD) or a flash memory card (FC) equipped on the computer device, etc., and the memory can also be a combination of an internal storage unit and an external storage device of a computer device. The memory is used to store computer programs and other programs and data required by the computer device. The memory can also be used to temporarily store data that has been output or is to be output, and this application does not impose any restrictions on this.

[0186] A person skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of protection of the present application is limited to these examples. In line with the concept of the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as above, which are not provided in detail for the sake of simplicity.

[0187] One or more embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the present application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of the present application should be included in the protection scope of the present application.

Claims

1. A power grid energy storage dispatching method based on deep learning and energy management, It is characterized in that Applied to a power distribution network, the power distribution network includes multiple power distribution areas, each of the power distribution areas includes multiple power loads and multiple microgrids built based on photovoltaic and wind power generation systems, and each of the power distribution areas is also equipped with a shared energy storage device, the shared energy storage device is used to store the surplus electric energy produced by all the microgrids in the power distribution area, the surplus electric energy represents the part of the electric energy produced by the microgrid in the same time period that exceeds the internal energy supply of the microgrid, the method comprises the following steps: For each of the power distribution areas, obtaining historical power consumption data and real-time power consumption data of all the power loads in the power distribution area; Based on a deep convolutional neural network and using the historical power consumption data, a power consumption feature recognition model is constructed, and the real-time power consumption data is analyzed using the power consumption feature recognition model to predict the power consumption features of the power load; Calculate the predicted total power consumption of the power distribution area according to the predicted power consumption characteristics of all the power loads; Acquire historical weather data and weather forecast data of the power distribution area, and historical surplus electric energy data of all the microgrids, 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 and using the historical weather data and the historical surplus electric energy data, a power generation prediction model is constructed, the weather prediction data is preprocessed, and the preprocessed weather prediction data is input into the power generation prediction model to obtain the predicted surplus electric energy of the microgrid; Combine the predicted total power consumption, the predicted surplus power of all the microgrids and the remaining stored power of the shared energy storage device to calculate the predicted energy storage data of the shared energy storage device, and store the predicted energy storage data in the energy storage server carried by the shared energy storage device; The energy storage server of each shared energy storage device builds an energy dispatching blockchain network for the power distribution network; Selecting a leader server among all the energy storage servers according to the server status of the energy storage servers; Mark the shared energy storage device whose predicted energy storage data is lower than a preset electric energy data threshold as a target shared energy storage device, and mark the energy storage server of the target shared energy storage device as a target energy storage server; Deploy a smart contract on the target energy storage server, automatically generate an energy dispatch request for the target shared energy storage device through the target energy storage server, and upload the energy dispatch request to the energy dispatch blockchain network, wherein the smart contract is used to define the generation rules and process of the energy dispatch request, and is used to automatically generate the energy dispatch request according to the predicted energy storage data and current status of the target shared energy storage device; Using the leading server to broadcast the electric energy dispatch request to other energy storage servers other than the target energy storage server; When the energy storage server receives the electric energy dispatch request, it generates electric energy dispatch information by combining the electric energy dispatch request and the predicted energy storage data stored locally, and uploads the electric energy dispatch information to the energy dispatch blockchain network; All the electric energy dispatching information is aggregated through the leading server and an electric energy dispatching plan for the power distribution network is generated.

2. According to the power grid energy storage scheduling method based on deep learning and energy management according to claim 1, It is characterized in that The method of constructing a power consumption feature recognition model based on a deep convolutional neural network and using the historical power consumption data, analyzing the real-time power consumption data using the power consumption feature recognition model, and predicting the predicted power consumption features of the power load comprises the following steps: Construct an initial electricity consumption feature recognition model based on a deep convolutional neural network; Preprocessing all the historical electricity consumption data into a historical electricity consumption data set, wherein all the historical electricity consumption data are pre-marked with corresponding historical electricity consumption features; Performing feature correlation analysis on the historical power consumption features of the historical power consumption data set to obtain a power consumption feature classification matrix of the historical power consumption data set; Using the power consumption feature classification matrix to train the initial power consumption feature recognition model to obtain a trained power consumption feature recognition model; The real-time power consumption data is analyzed using the power consumption feature recognition model to predict the power consumption features of the power load.

3. According to the power grid energy storage scheduling method based on deep learning and energy management according to claim 2, It is characterized in that The performing feature correlation analysis on the historical power consumption features of the historical power consumption data set to obtain the power consumption feature classification matrix of the historical power consumption data set comprises the following steps: Extracting the historical electricity consumption features of the historical electricity consumption data set; The connection function is used to calculate the power consumption feature correlation of the historical power consumption features of different feature types; The power consumption feature correlation is analyzed using a decision tree learning algorithm based on information entropy, and a power consumption feature classification matrix of the historical power consumption data set is obtained.

4. According to the power grid energy storage scheduling method based on deep learning and energy management according to claim 3, It is characterized in that The step of using the power consumption feature classification matrix to train the initial power consumption feature recognition model to obtain the trained power consumption feature recognition model comprises the following steps: Using a particle swarm optimization algorithm to select the optimal power consumption feature from the power consumption feature classification matrix; Based on the optimal electricity consumption characteristics and using a weighted least squares method, an optimal characteristic matrix is ​​constructed; Inputting the optimal feature matrix into the initial power consumption feature recognition model, and calculating the bias item parameters of all fully connected layers in the initial power consumption feature recognition model; Calculating the parameter error of the bias term parameter by using the preset optimal parameter; If the parameter error is greater than or equal to the preset error threshold, re-execute all the above steps until the parameter error is less than the error threshold; If the parameter error is less than the error threshold, the model training process of the initial power consumption feature recognition model is completed to obtain a trained power consumption feature recognition model.

5. The power grid energy storage dispatching method based on deep learning and energy management according to claim 4, It is characterized in that The method of selecting the optimal power consumption feature from the power consumption feature classification matrix using a particle swarm optimization algorithm comprises the following steps: Randomly selecting initial electricity usage features from the electricity usage feature classification matrix; Initializing the initial power consumption characteristics into two-dimensional characteristic data using preset particle swarm algorithm parameters, and constructing an initial particle swarm according to the two-dimensional characteristic data; The initial particle swarm is optimized multiple times by using a particle swarm optimization algorithm until the initial particle swarm is optimized to be an optimal particle swarm; All the initial power usage characteristics in the optimal particle group are used as optimal power usage characteristics.

6. The power grid energy storage dispatching method based on deep learning and energy management according to claim 1, It is characterized in that The method of constructing a power generation prediction model based on a support vector machine and using the historical weather data and the historical surplus electric 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 electric energy of the microgrid comprises the following steps: Constructing an initial power generation prediction model based on support vector machine; Using a grey correlation algorithm to respectively calculate the data correlation between different types of the historical weather data and the historical surplus electric energy data; Integrate all the historical weather data whose data relevance is greater than a preset relevance threshold into a historical weather data set; Iteratively training the initial power generation prediction model in combination with the historical weather data set and the historical surplus electric energy data until the model parameters in the initial power generation prediction model are trained to the optimal model parameters, thereby obtaining a trained power generation prediction model; The weather forecast data is preprocessed, and the preprocessed weather forecast data is input into the power generation prediction model to obtain the predicted surplus electric energy of the microgrid.

7. The power grid energy storage dispatching method based on deep learning and energy management according to claim 6, It is characterized in that The preprocessing of the weather forecast data and inputting the preprocessed weather forecast data into the power generation prediction model to obtain the predicted surplus electric energy of the microgrid comprises the following steps: Extract all weather data types from the historical weather data set; Screening out invalid weather forecast data whose data type is different from all weather data types in the weather forecast data; All the remaining weather forecast data are input into the power generation forecast model to obtain the predicted surplus electric energy of the microgrid.

8. The power grid energy storage dispatching method based on deep learning and energy management according to claim 1, It is characterized in that The process of aggregating all the electric energy dispatching information and generating the electric energy dispatching plan for the power distribution network through the leading server comprises the following steps: Aggregating all the electric energy dispatching information and generating an initial electric energy dispatching plan for the power distribution network through the leading server; Broadcasting the initial power dispatching plan to all the energy storage servers through the leading server; Using the consensus mechanism to verify whether the initial power dispatching plan has been verified; If the initial electric energy dispatching scheme fails to be verified, the electric energy dispatching information regenerated by all the energy storage servers is obtained through the energy dispatching blockchain network, and the above scheme generation step and scheme verification step are repeated until the initial electric energy dispatching scheme is verified; If the initial power dispatching plan passes the verification, the initial power dispatching plan will be used as the power dispatching plan for the power distribution network.

9. A power grid energy storage dispatching system based on deep learning and energy management, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer program, it implements the power grid energy storage scheduling method based on deep learning and energy management as described in any one of claims 1 to 8.

Citation Information

Patent Citations

  • Micro-grid energy and load scheduling method, device, equipment, medium and product

    CN118137509A

  • Shared energy storage scheduling optimization method and device, medium and product

    CN118627680A