Control Method for Distributed Photovoltaic Grid-Connected Power Generation and Energy Storage Equipment Based on Big Data

Abstract

The present invention relates to a control method for distributed photovoltaic grid-connected power generation and energy storage devices based on big data, which includes: obtaining an attention matrix for each power interaction index based on the multi-head attention mechanism, and connecting the attention matrices of all power interaction indexes using a feed-forward neural network to generate an interaction attention matrix; determining deep feature vectors and anomaly correlation degrees for each power interaction index based on the interaction attention matrix, and sorting the deep feature vectors of all power interaction indexes according to the anomaly correlation degree to obtain a deep feature vector sequence; performing multi-layer perception processing on the deep feature vector sequence, the first power expansion feature, the second power expansion feature, and the third power expansion feature to obtain a first anomaly perception feature and a second anomaly perception feature; and inputting the first anomaly perception feature and the second anomaly perception feature into a pre-trained power grid monitoring model to determine whether an anomaly occurs in the photovoltaic power grid.

Description

Technical Field

[0001] The present invention relates to the fields of big data and distributed photovoltaic grid connection, and particularly to a control method for distributed photovoltaic grid-connected power generation and energy storage equipment based on big data. Background Art

[0002] Distributed photovoltaic power generation is of great significance for optimizing the energy structure, promoting energy conservation and emission reduction, and realizing sustainable economic development. Due to the characteristics of low voltage level, small capacity, wide distribution, and local consumption of distributed photovoltaics, there are many ways of electricity theft in distributed photovoltaics, and the difficulty of realizing electricity theft is low.

[0003] Monitoring electricity theft in distributed photovoltaic power generation involves various data information. The patent publication number (CN113341216A) "Anti-electricity-theft method for distributed photovoltaic power generation system" discloses judging whether there is an electricity theft behavior by comparing the total electric energy output from the photovoltaic parallel structure to the inverter with the total electric energy output from the inverter to the AC power grid. However, since the photovoltaic grid connection connects all user terminals, there will be a situation where illegal users steal the power generation of other users. When illegal users steal the power generation of other users, it is impossible to identify the electricity theft behavior only by comparing the total electric energy.

[0004] With the development of distributed photovoltaic power generation, how to design a method to identify the behavior of illegal user terminals stealing the power generation of other user terminals through the power interaction data between user terminals and the power interaction data between user terminals and the central control terminal, and quickly identify abnormal user terminals is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a control method for distributed photovoltaic grid-connected power generation and energy storage equipment based on big data, which includes: several user terminals are connected to a central control energy storage device through a distributed network to generate a photovoltaic power grid; each user terminal includes a photovoltaic power generation device and a terminal energy storage device;

[0006] Obtain the first power interaction data of each photovoltaic power generation device in the photovoltaic power grid, perform multi-dimensional feature interaction on the first power interaction data of all photovoltaic power generation devices in the photovoltaic power grid to obtain the first power interaction feature of the photovoltaic power grid, and then perform feature expansion on the first power interaction feature to obtain the first interaction expansion feature of the photovoltaic power grid;

[0007] Obtain the second power interaction data of each terminal energy storage device in the photovoltaic power grid, perform multi-dimensional feature interaction on the second power interaction data of all terminal energy storage devices in the photovoltaic power grid to obtain the second power interaction feature of the photovoltaic power grid, and then perform feature expansion on the second power interaction feature to obtain the second interaction expansion feature of the photovoltaic power grid;

[0008] Obtain the third power interaction data of the central control energy storage device of the photovoltaic power grid, extract the data features of the third power interaction data to generate the third power interaction feature, and then perform feature expansion on the third power interaction feature to obtain the third interaction expansion feature of the photovoltaic power grid;

[0009] Obtain a number of power interaction indicators from the database, obtain the attention matrix of each power interaction indicator based on the multi-head attention mechanism, and then use a feed-forward neural network to connect the attention matrices of all power interaction indicators to generate an interaction attention matrix;

[0010] Based on the interaction attention matrix, determine the deep feature vector and anomaly correlation degree of each power interaction indicator, and sort the deep feature vectors of all power interaction indicators according to the anomaly correlation degree to obtain a deep feature vector sequence;

[0011] Perform multi-layer perception processing on the deep feature vector sequence, the first power expansion feature, the second power expansion feature, and the third power expansion feature to obtain the first anomaly perception feature and the second anomaly perception feature;

[0012] Input the first anomaly perception feature and the second anomaly perception feature into a pre-trained power grid monitoring model to determine whether the photovoltaic power grid has an anomaly, and when the photovoltaic power grid has an anomaly, identify the abnormal user end, and then block all operations of the photovoltaic power generation equipment and the terminal energy storage equipment of the abnormal user end.

[0013] According to a preferred embodiment, the first power interaction data is the power interaction data between photovoltaic power generation devices; the second power interaction data is the power interaction data between terminal energy storage devices; the third power interaction device is the power interaction data between the central control energy storage device and the terminal energy storage device.

[0014] According to a preferred embodiment, the power interaction data is voltage input / output data, current input / output data, and power input / output data.

[0015] According to a preferred embodiment, the photovoltaic power generation device is a device that can directly convert light energy into electrical energy and is composed of a solar panel, a controller, and an inverter;

[0016] The terminal energy storage device is used to store the electrical energy generated by the photovoltaic power generation device of the same user end or deliver electrical energy to other user ends;

[0017] The central control energy storage device is used to store the electrical energy generated by the photovoltaic power generation devices of all user ends or deliver electrical energy to all user ends.

[0018] According to a preferred embodiment, the photovoltaic power grid is a power grid composed of a plurality of photovoltaic power generation devices, a plurality of terminal energy storage devices, and a central control energy storage device; the photovoltaic power grid has the functions of producing electric energy, transmitting electric energy, and distributing electric energy.

[0019] According to a preferred embodiment, performing multi-layer perception processing on the deep feature vector sequence, the first power expansion feature, the second power expansion feature, and the third power expansion feature to obtain the first anomaly perception feature and the second anomaly perception feature includes:

[0020] Performing feature fusion on the deep feature vector sequence with the first power expansion feature, the second power expansion feature, and the third power expansion feature respectively to generate a first deep expansion feature sequence, a second deep expansion feature sequence, and a third deep expansion feature sequence;

[0021] Inputting the first deep expansion feature sequence and the second deep expansion feature sequence into a multi-layer perceptron for training to output the first anomaly perception feature;

[0022] Inputting the third deep expansion feature sequence into a multi-layer perceptron for training to output the second anomaly perception feature.

[0023] According to a preferred embodiment, the power interaction index is a relevant power index for judging whether the photovoltaic power grid has an anomaly, and it includes: power generation amount, transmitted power amount, and stored power amount.

[0024] According to a preferred embodiment, obtaining the attention matrix of each power interaction index based on the multi-head attention mechanism includes:

[0025] Traversing all the power interaction indexes, taking the power interaction index being traversed as the target interaction index, and then extracting the index feature vectors of all the power interaction indexes;

[0026] Taking the dot product of the index feature vector of the target interaction index and the index feature vectors of other power interaction indexes as the index correlation degree between the target interaction index and other power interaction indexes, and enhancing the numerical stability of the index correlation degree between the target interaction index and other power interaction indexes through a preset proportionality coefficient;

[0027] Performing linear projection on the index feature vectors of the target interaction index and other each power interaction index based on the index correlation degree between the target interaction index and other each power interaction index to generate a plurality of linear interaction vectors, performing attention pooling on all the linear interaction vectors, and then splicing all the linearly interaction vectors after attention pooling to obtain the attention matrix of the target interaction index;

[0028] Repeating the above steps to obtain the attention matrix of each power interaction index.

[0029] According to a preferred embodiment, performing multi-dimensional feature interaction on the first power interaction data of all photovoltaic power generation devices in the photovoltaic grid to obtain the first power interaction feature of the photovoltaic grid includes:

[0030] Extracting data features of the first power interaction data of each photovoltaic power generation device to obtain a plurality of power interaction features, and inputting all the power interaction features into a convolutional neural network to generate a plurality of interaction convolution features;

[0031] All interactive convolution features are subjected to feature fusion to obtain interactive fusion features, and the interactive fusion features are input into a deconvolution neural network to output a first power interaction feature.

[0032] According to a preferred embodiment, performing feature expansion on the first power interaction feature to obtain a first interaction expansion feature of the photovoltaic network includes:

[0033] Decomposing the first power interaction feature to obtain a plurality of first attribute features, and obtaining a distribution feature of each first attribute feature;

[0034] Determine the characteristic value of each first attribute feature from top to bottom according to the tree arrangement order of each first attribute feature by using the roulette algorithm and the distribution characteristics of each first attribute feature;

[0035] Analyzing the distribution characteristics of each first attribute feature using a multiple linear regression algorithm to construct a multiple linear regression model for each first attribute feature;

[0036] Inputting the characteristic value of each first attribute feature into the corresponding multiple linear regression model to output the cross coefficient of each first attribute feature;

[0037] A plurality of cross-attribute features are generated through cross-coefficients of all first attribute features and all first attribute features, and the first power interaction feature is feature expanded according to the cross-attribute features to obtain a first interaction expansion feature.

[0038] The present invention has the following beneficial effects: This application analyzes the power interaction behavior between user terminals and user terminals, and between user terminals and central control terminals through the power interaction data between photovoltaic power generation equipment and photovoltaic power generation equipment, the power interaction data between terminal energy storage equipment and terminal energy storage equipment, and the power interaction data between central control energy storage equipment and terminal energy storage equipment, thereby identifying whether there is any illegal user terminal stealing electricity from other user terminals, thereby avoiding economic losses for other user terminals.

[0039] In addition, a multiple regression model is established based on the distribution characteristics of attribute features, and the cross-relationship between attribute features is obtained according to the multiple regression model, so as to explore the mutual influence relationship between features, thereby realizing the feature expansion of power interaction features, increasing the feature dimension, and improving the accuracy of subsequent power grid anomaly recognition.

[0040] On the other hand, the attention matrix of each power interaction index is obtained through the multi-head attention mechanism, and the interaction attention matrix is generated according to the attention matrices of all power interaction indexes, so as to quickly and accurately locate the most sensitive relevant features for identifying photovoltaic power grid anomalies in the power interaction indexes, reducing the waste of computing resources while improving the accuracy and recognition efficiency of photovoltaic power grid anomaly recognition. Brief Description of the Drawings

[0041] Figure 1 It is a flowchart of a control method for distributed photovoltaic grid-connected power generation and energy storage equipment based on big data provided for an exemplary embodiment. Detailed Embodiments

[0042] To make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with the specific embodiments and with reference to the accompanying drawings. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of the present invention. In addition, in the following description, the descriptions of well-known structures and technologies are omitted to avoid unnecessarily confusing the concepts of the present invention.

[0043] The terms used in the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "a", "the" and "said" used in the present invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0044] It should be understood that although the terms first, second, third, etc. may be used in the present invention to describe various information, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present invention, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "when" or "while" or "in response to a determination".

[0045] See Figure 1 , in one embodiment, the control method for distributed photovoltaic grid-connected power generation and energy storage equipment based on big data includes the following steps:

[0046] S1. Several clients are connected to a central control energy storage device through a distributed network to generate a photovoltaic power grid; each client includes a photovoltaic power generation device and a terminal energy storage device.

[0047] Optionally, the photovoltaic power generation device is a device that can directly convert light energy into electrical energy and is composed of a solar panel, a controller, and an inverter; the terminal energy storage device is used to store the electrical energy generated by the photovoltaic power generation device of the same client or transmit electrical energy to other clients; the central control energy storage device is used to store the electrical energy generated by the photovoltaic power generation devices of all clients or transmit electrical energy to all clients.

[0048] Optionally, the photovoltaic power grid is a power grid composed of several photovoltaic power generation devices, several terminal energy storage devices, and a central control energy storage device; the photovoltaic power grid has the functions of producing electrical energy, transmitting electrical energy, and distributing electrical energy.

[0049] Optionally, one client includes several photovoltaic power generation devices and several terminal energy storage devices, and each photovoltaic power generation device and terminal energy storage device has a device number, which is used to uniquely identify the photovoltaic power generation device and the terminal energy storage device, and record which client the photovoltaic power generation device and the terminal energy storage device belong to.

[0050] S2. Obtain the first power interaction data of each photovoltaic power generation device in the photovoltaic power grid, perform multi-dimensional feature interaction on the first power interaction data of all photovoltaic power generation devices in the photovoltaic power grid to obtain the first power interaction feature of the photovoltaic power grid, and then perform feature expansion on the first power interaction feature to obtain the first interaction expansion feature of the photovoltaic power grid.

[0051] Optionally, the first power interaction data is the power interaction data between photovoltaic power generation devices; the power interaction data is voltage input / output data, current input / output data, and power input / output data.

[0052] In one embodiment, performing multi-dimensional feature interaction on the first power interaction data of all photovoltaic power generation devices in the photovoltaic power grid to obtain the first power interaction feature of the photovoltaic power grid includes:

[0053] Extract the data features of the first power interaction data of each photovoltaic power generation device to obtain several power interaction features, and input all the power interaction features into a convolutional neural network to generate several interaction convolutional features;

[0054] Perform feature fusion on all the interaction convolutional features to obtain an interaction fusion feature, and input the interaction fusion feature into a deconvolutional neural network to output the first power interaction feature.

[0055] In one embodiment, performing feature expansion on the first power interaction feature to obtain the first interaction expansion feature of the photovoltaic power grid includes:

[0056] Decomposing the first power interaction feature to obtain a plurality of first attribute features, and obtaining a distribution feature of each first attribute feature;

[0057] Determine the characteristic value of each first attribute feature from top to bottom according to the tree arrangement order of each first attribute feature using the roulette algorithm and the distribution characteristics of each first attribute feature;

[0058] Analyzing the distribution characteristics of each first attribute feature using a multiple linear regression algorithm to construct a multiple linear regression model for each first attribute feature;

[0059] Inputting the characteristic value of each first attribute feature into the corresponding multiple linear regression model to output the cross coefficient of each first attribute feature;

[0060] A plurality of cross-attribute features are generated through cross-coefficients of all first attribute features and all first attribute features, and the first power interaction feature is feature expanded according to the cross-attribute features to obtain a first interaction expansion feature.

[0061] In this embodiment, a multivariate regression model is established based on the distribution characteristics of the first attribute features, and the cross-relationships between the first attribute features are obtained based on the multivariate regression model, and the mutual influence relationship between features is explored, thereby achieving feature expansion of the first power interaction feature, increasing the feature dimension, and improving the accuracy of subsequent power grid anomaly identification.

[0062] S3. Obtain the second power interaction data of each terminal energy storage device in the photovoltaic grid, perform multi-dimensional feature interaction on the second power interaction data of all terminal energy storage devices in the photovoltaic grid to obtain a second power interaction feature of the photovoltaic grid, and then perform feature expansion on the second power interaction feature to obtain a second interaction expansion feature of the photovoltaic grid.

[0063] Optionally, the second power interaction data is power interaction data between terminal energy storage devices; the power interaction data is voltage input and output data, current input and output data, and power input and output data.

[0064] In one embodiment, performing multi-dimensional feature interaction on the second power interaction data of all terminal energy storage devices in the photovoltaic power grid to obtain the second power interaction feature of the photovoltaic power grid includes:

[0065] Extracting data features of the second power interaction data of each terminal energy storage device to obtain a plurality of power interaction features, and inputting all the power interaction features into a convolutional neural network to generate a plurality of interaction convolution features;

[0066] All interactive convolution features are fused to obtain interactive fusion features, and the interactive fusion features are input into a deconvolution neural network to output a second power interaction feature.

[0067] In one embodiment, performing feature expansion on the second power interaction feature to obtain the second interaction expansion feature of the photovoltaic power grid includes:

[0068] Decomposing the second power interaction feature to obtain a plurality of second attribute features, and obtaining a distribution feature of each second attribute feature;

[0069] Determine the characteristic value of each second attribute feature from top to bottom according to the tree arrangement order of each second attribute feature by using the roulette algorithm and the distribution characteristics of each second attribute feature;

[0070] Analyzing the distribution characteristics of each second attribute feature using a multiple linear regression algorithm to construct a multiple linear regression model for each second attribute feature;

[0071] Inputting the characteristic value of each second attribute feature into the corresponding multiple linear regression model to output the cross coefficient of each second attribute feature;

[0072] A plurality of cross-attribute features are generated through cross-coefficients of all second attribute features and all second attribute features, and the second power interaction feature is feature expanded according to the cross-attribute features to obtain a second interaction expansion feature.

[0073] In this embodiment, a multivariate regression model is established based on the distribution characteristics of the second attribute features, and the cross-relationships between the second attribute features are obtained based on the multivariate regression model, and the mutual influence relationship between features is explored, thereby achieving feature expansion of the second power interaction feature, increasing the feature dimension, and improving the accuracy of subsequent power grid anomaly identification.

[0074] S4. Obtain third power interaction data of the central control energy storage device of the photovoltaic power grid, extract data features of the third power interaction data to generate a third power interaction feature, and then perform feature expansion on the third power interaction feature to obtain a third interaction expansion feature of the photovoltaic power grid.

[0075] Optionally, the third power interaction device is power interaction data between the central control energy storage device and the terminal energy storage device; the power interaction data is voltage input and output data, current input and output data, and power input and output data.

[0076] In one embodiment, performing feature expansion on the third power interaction feature to obtain the third interaction expansion feature of the photovoltaic power grid includes:

[0077] Decomposing the third power interaction feature to obtain a plurality of third attribute features, and obtaining a distribution feature of each third attribute feature;

[0078] Determine the characteristic value of each third attribute feature from top to bottom according to the tree arrangement order of each third attribute feature by using the roulette algorithm and the distribution characteristics of each third attribute feature;

[0079] The distribution characteristics of each third attribute feature are analyzed using a multiple linear regression algorithm to construct a multiple linear regression model for each third attribute feature;

[0080] Inputting the characteristic value of each third attribute feature into the corresponding multiple linear regression model to output the cross coefficient of each third attribute feature;

[0081] A plurality of cross-attribute features are generated through cross-coefficients of all third attribute features and all third attribute features, and the third power interaction feature is feature expanded according to the cross-attribute features to obtain a third interaction expansion feature.

[0082] In this embodiment, a multivariate regression model is established based on the distribution characteristics of the third attribute features, and the cross-relationships between the third attribute features are obtained based on the multivariate regression model, and the mutual influence relationship between features is explored, thereby achieving feature expansion of the third power interaction feature, increasing the feature dimension, and improving the accuracy of subsequent power grid anomaly identification.

[0083] S5. Obtain several power interaction indicators from the database, and obtain the attention matrix of each power interaction indicator based on the multi-head attention mechanism, and then use the feedforward neural network to connect the attention matrices of all power interaction indicators to generate an interaction attention matrix.

[0084] Optionally, the power interaction index is a related power index used to determine whether an abnormality occurs in the photovoltaic power grid, which includes: power generation, power transmission and storage.

[0085] In one embodiment, obtaining an attention matrix for each power interaction indicator based on a multi-head attention mechanism includes:

[0086] Traversing all power interaction indicators, taking the power interaction indicator being traversed as the target interaction indicator, and then extracting the indicator feature vectors of all power interaction indicators; the indicator feature vectors are used to characterize the relevant features of the power interaction indicators;

[0087] The dot product of the index feature vector of the target interaction index and the index feature vectors of other power interaction indexes is used as the index correlation degree between the target interaction index and other power interaction indexes, and the proportionality coefficient is used to enhance the numerical stability of the index correlation degree between the target interaction index and other power interaction indexes; the index correlation degree is used to characterize the degree of mutual influence between power interaction indexes;

[0088] Based on the index correlation degree between the target interaction index and each other power interaction index, linear projection is performed on the index feature vectors of the target interaction index and each other power interaction index to generate a number of linear interaction vectors, and attention pooling is performed on all the linear interaction vectors, and then all the linearly interaction vectors after attention pooling are concatenated to obtain the attention matrix of the target interaction index;

[0089] Repeat the above steps to obtain the attention matrix of each power interaction index.

[0090] In this embodiment, the attention matrix of each power interaction index is obtained through the multi-head attention mechanism, and the interactive attention matrix is generated according to the attention matrices of all power interaction indexes, so as to quickly and accurately locate the most sensitive correlation features for identifying photovoltaic grid anomalies in power interaction indexes, while improving the accuracy and recognition efficiency of photovoltaic grid anomaly recognition and reducing the waste of computing resources.

[0091] S6. Determine the deep feature vector and anomaly correlation degree of each power interaction index based on the interactive attention matrix, and sort the deep feature vectors of all power interaction indexes according to the anomaly correlation degree to obtain a deep feature vector sequence.

[0092] Optionally, the anomaly correlation degree of the power interaction index is the probability that the power interaction index will show an anomaly when the photovoltaic grid shows an anomaly.

[0093] Optionally, the deep feature vector is used to characterize the deep features of the power interaction index.

[0094] S7. Perform multi-layer perception processing on the deep feature vector sequence, the first power expansion feature, the second power expansion feature, and the third power expansion feature to obtain a first anomaly perception feature and a second anomaly perception feature.

[0095] In one embodiment, performing multi-layer perception processing on the deep feature vector sequence, the first power expansion feature, the second power expansion feature, and the third power expansion feature to obtain a first anomaly perception feature and a second anomaly perception feature includes:

[0096] Respectively perform feature fusion on the deep feature vector sequences with the first power expansion feature, the second power expansion feature, and the third power expansion feature to generate a first deep expansion feature sequence, a second deep expansion feature sequence, and a third deep expansion feature sequence;

[0097] Input the first deep expansion feature sequence and the second deep expansion feature sequence into a multi-layer perceptron for training to output a first anomaly perception feature;

[0098] Input the third deep expansion feature sequence into a multi-layer perceptron for training to output a second anomaly perception feature.

[0099] Optionally, the multi-layer perceptron includes an input layer, a hidden layer, and an output layer.

[0100] In one embodiment, inputting the first deep expansion feature sequence and the second deep expansion feature sequence into a multi-layer perceptron for training to output a first anomaly perception feature includes:

[0101] Fully connect the first deep expansion feature sequence and the second deep expansion feature sequence through the input layer to generate a deep fully connected feature sequence;

[0102] Pass the deep fully connected feature sequence through multiple hidden layers and activation functions to generate a number of anomaly features;

[0103] Fully connect all the anomaly features through the output layer to output the first anomaly perception feature.

[0104] In one embodiment, inputting the third deep expansion feature sequence into a multi-layer perceptron for training to output a second anomaly perception feature includes:

[0105] Input the third deep expansion feature sequence into the multi-layer perceptron through the input layer;

[0106] Pass the third deep expansion feature sequence through multiple hidden layers and activation functions to generate a number of anomaly features;

[0107] Fully connect all the anomaly features through the output layer to output the second anomaly perception feature.

[0108] S8. Input the first anomaly perception feature and the second anomaly perception feature into a pre-trained power grid monitoring model to determine whether the photovoltaic power grid is abnormal. When the photovoltaic power grid is abnormal, identify the abnormal user terminals, and then block all operations of the photovoltaic power generation devices and terminal energy storage devices of the abnormal user terminals.

[0109] Optionally, obtain model training data from a database and train a power grid monitoring model based on the model training data.

[0110] In one embodiment, the first anomaly perception feature and the second anomaly perception feature are input into a pre-trained power grid monitoring model to quickly and accurately determine whether there is an anomaly in the photovoltaic power grid. When an anomaly occurs in the photovoltaic power grid, for example, when the power generation amount, the stored power amount, and the used power amount do not match, the abnormal user end can be quickly and accurately identified. There is a possibility of electricity theft at this abnormal user end. At this time, all operations of all photovoltaic power generation devices and terminal energy storage devices included in this abnormal user end will be blocked to avoid further losses.

[0111] This application analyzes the power interaction behaviors between user ends and between user ends and the central control end by analyzing the power interaction data between photovoltaic power generation devices, the power interaction data between terminal energy storage devices, and the power interaction data between the central control energy storage device and the terminal energy storage device, so as to identify whether there is a behavior of an illegal user end stealing the electricity of other user ends, thereby avoiding the economic losses of other user ends.

[0112] In addition, a multiple regression model is established based on the distribution characteristics of the attribute features, and the cross relationship between the attribute features is obtained according to the multiple regression model, and the mutual influence relationship between the features is mined, so as to realize the feature expansion of the power interaction features, increase the feature dimension, and improve the accuracy of subsequent power grid anomaly identification.

[0113] On the other hand, the attention matrix of each power interaction index is obtained through the multi-head attention mechanism, and the interactive attention matrix is generated according to the attention matrices of all power interaction indexes, so as to quickly and accurately locate the most sensitive relevant features for identifying photovoltaic power grid anomalies in the power interaction indexes, while improving the accuracy and identification efficiency of photovoltaic power grid anomaly identification and reducing the waste of computing resources.

[0114] In addition, in each embodiment of this article, each functional unit can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0115] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution herein, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments herein. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs.

[0116] Specific embodiments are applied in this article to elaborate on the principles and implementation manners of this article. The description of the above embodiments is only used to help understand the method and its core idea herein; at the same time, for those of ordinary skill in the art, according to the idea herein, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to this article.

Claims

1. A control method for a distributed photovoltaic grid-connected power generation and energy storage device based on big data, characterized in that Several clients are connected to the central control energy storage device through a distributed network to generate a photovoltaic power grid; each client includes a photovoltaic power generation device and a terminal energy storage device; Obtain the first power interaction data of each photovoltaic power generation device in the photovoltaic power grid, perform multi-dimensional feature interaction on the first power interaction data of all photovoltaic power generation devices in the photovoltaic power grid to obtain the first power interaction feature of the photovoltaic power grid, and then perform feature expansion on the first power interaction feature to obtain the first interaction expansion feature of the photovoltaic power grid; Obtain the second power interaction data of each terminal energy storage device in the photovoltaic power grid, perform multi-dimensional feature interaction on the second power interaction data of all terminal energy storage devices in the photovoltaic power grid to obtain the second power interaction feature of the photovoltaic power grid, and then perform feature expansion on the second power interaction feature to obtain the second interaction expansion feature of the photovoltaic power grid; Obtain the third power interaction data of the central control energy storage device of the photovoltaic power grid, extract the data features of the third power interaction data to generate the third power interaction feature, and then perform feature expansion on the third power interaction feature to obtain the third interaction expansion feature of the photovoltaic power grid; Obtain several power interaction indicators from the database, obtain the attention matrix of each power interaction indicator based on the multi-head attention mechanism, and then use the feed-forward neural network to connect the attention matrices of all power interaction indicators to generate an interaction attention matrix; Determine the deep feature vector and anomaly correlation degree of each power interaction indicator based on the interaction attention matrix, and sort the deep feature vectors of all power interaction indicators according to the anomaly correlation degree to obtain a deep feature vector sequence; Perform multi-layer perception processing on the deep feature vector sequence, the first power expansion feature, the second power expansion feature, and the third power expansion feature to obtain the first anomaly perception feature and the second anomaly perception feature; Input the first anomaly perception feature and the second anomaly perception feature into a pre-trained power grid monitoring model to determine whether the photovoltaic power grid has an anomaly, and identify the abnormal client when the photovoltaic power grid has an anomaly, and then block all operations of the photovoltaic power generation device and the terminal energy storage device of the abnormal client.

2. The method according to claim 1, characterized in that The first power interaction data is the power interaction data between photovoltaic power generation devices; the second power interaction data is the power interaction data between terminal energy storage devices; the third power interaction data is the power interaction data between the central control energy storage device and the terminal energy storage device.

3. The method according to claim 2, wherein The power interaction data is voltage input / output data, current input / output data, and power input / output data.

4. The method according to claim 3, wherein The photovoltaic power generation device is a device that can directly convert light energy into electrical energy and is composed of a solar panel, a controller, and an inverter; The terminal energy storage device is used to store the electrical energy generated by the photovoltaic power generation device of the same client or deliver electrical energy to other clients; The central control energy storage device is used to store the electrical energy generated by the photovoltaic power generation devices of all clients or deliver electrical energy to all clients.

5. The method according to claim 4, wherein The photovoltaic power grid is a power grid composed of several photovoltaic power generation devices, several terminal energy storage devices, and a central control energy storage device; the photovoltaic power grid has the functions of generating electrical energy, transmitting electrical energy, and distributing electrical energy.

6. The method according to claim 5, wherein Performing multi-layer perception processing on the deep feature vector sequence, the first power expansion feature, the second power expansion feature, and the third power expansion feature to obtain the first abnormality perception feature and the second abnormality perception feature includes: Fusing the deep feature vector sequence with the first power expansion feature, the second power expansion feature, and the third power expansion feature to generate a first deep expansion feature sequence, a second deep expansion feature sequence, and a third deep expansion feature sequence; Inputting the first deep-level expanded feature sequence and the second deep-level expanded feature sequence into a multi-layer perceptron for training to output a first abnormality perception feature; The third deep-level expanded feature sequence is input into a multi-layer perceptron for training to output a second abnormality perception feature.

7. The method according to claim 6, wherein The power interaction index is a related power index used to determine whether an abnormality occurs in the photovoltaic power grid, which includes: power generation, power transmission and storage.

8. The method according to claim 7, characterized in that, The attention matrix for each power interaction indicator obtained based on the multi-head attention mechanism includes: Traverse all power interaction indicators, take the power interaction indicator being traversed as the target interaction indicator, and then extract the indicator feature vectors of all power interaction indicators; The dot product of the indicator feature vector of the target interaction indicator and the indicator feature vector of other power interaction indicators is used as the indicator correlation between the target interaction indicator and the other power interaction indicators, and the numerical stability of the indicator correlation between the target interaction indicator and the other power interaction indicators is enhanced by a preset proportional coefficient; Based on the correlation between the target interaction indicator and each other power interaction indicator, the indicator feature vectors of the target interaction indicator and each other power interaction indicator are linearly projected to generate several linear interaction vectors. All linear interaction vectors are then attention pooled. All linear interaction vectors after attention pooling are then concatenated to obtain the attention matrix of the target interaction indicator. Repeat the above steps to obtain the attention matrix for each power interaction indicator.

9. The method according to claim 8, wherein Performing multi-dimensional feature interaction on the first power interaction data of all photovoltaic power generation devices in the photovoltaic grid to obtain the first power interaction feature of the photovoltaic grid includes: Extracting data features of the first power interaction data of each photovoltaic power generation device to obtain a plurality of power interaction features, and inputting all the power interaction features into a convolutional neural network to generate a plurality of interaction convolution features; All interactive convolution features are subjected to feature fusion to obtain interactive fusion features, and the interactive fusion features are input into a deconvolution neural network to output a first power interaction feature.

10. The method according to claim 9, characterized in that, The first interaction expansion feature of the photovoltaic power grid obtained by performing feature expansion on the first power interaction feature includes: Decomposing the first power interaction feature to obtain a plurality of first attribute features, and obtaining a distribution feature of each first attribute feature; Determine the characteristic value of each first attribute feature from top to bottom according to the tree arrangement order of each first attribute feature using the roulette algorithm and the distribution characteristics of each first attribute feature; Analyzing the distribution characteristics of each first attribute feature using a multiple linear regression algorithm to construct a multiple linear regression model for each first attribute feature; Input the eigenvalue of each first attribute feature into the corresponding multiple linear regression model to output the cross coefficient of each first attribute feature; Generate a number of cross-attribute features through the cross coefficients of all first attribute features and all first attribute features, and perform feature augmentation on the first power interaction feature according to the cross-attribute features to obtain the first interaction augmentation feature.

Images