Virtual power plant distributed data anomaly diagnosis method based on deep neural network

Abstract

The application provides a virtual power plant distributed data anomaly diagnosis method based on a deep neural network, relates to the technical field of artificial intelligence, and comprises the following steps: obtaining multi-modal data of each distributed energy of a virtual power plant to obtain processed distributed data; modeling each node of the virtual power plant and the relationship thereof into a graph structure to obtain a multi-time scale space-time feature vector; pre-training by constructing a self-supervised learning task to extract effective features between different time scales and modal data; generating a normal data distribution model by using a generative adversarial network, and integrating abnormal data samples into a comprehensive data set; obtaining a multi-task optimization model through shared features between tasks; obtaining a final optimized abnormal response system, and then completing the task of virtual power plant distributed data anomaly diagnosis through the abnormal response system. The application realizes efficient anomaly diagnosis and response processing of virtual power plant distributed data, and has significant technical advantages and practical application value.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a virtual power plant distributed data anomaly diagnosis method based on a deep neural network. BACKGROUND

[0002] With the continuous optimization of energy structure and the rapid development of renewable energy, virtual power plants as a new type of power system dispatching and management mode have received widespread attention. Virtual power plants can improve energy utilization efficiency and system stability by integrating distributed energy resources such as wind power, photovoltaic, and energy storage devices for unified scheduling and optimized management.

[0003] However, there are many challenges in the operation of existing virtual power plants, especially for distributed data anomaly detection and response processing, which rely on traditional monitoring and alarm systems. These systems have the following problems: high data complexity: the data volume of distributed energy is large and complex, and traditional methods are difficult to handle multi-modal and heterogeneous data; high real-time requirement: the power system requires real-time monitoring and rapid response, and existing technologies are difficult to meet the high real-time requirement; low detection accuracy: traditional anomaly detection techniques mostly use simple threshold judgment or experience-based rules, which are difficult to accurately capture complex abnormal patterns. Existing technologies generally set simple alarm thresholds or use experience-based rules to detect anomalies. These methods are not up to the task when faced with massive and complex data, and are prone to false positives and false negatives, affecting the safety and reliability of the power system. Therefore, there is an urgent need for a more intelligent and efficient anomaly diagnosis method to solve this problem. SUMMARY

[0004] The purpose of the present application is to provide a virtual power plant distributed data anomaly diagnosis method based on a deep neural network to improve the above problems. In order to achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0005] In a first aspect, the present application provides a virtual power plant distributed data anomaly diagnosis method based on a deep neural network, comprising:

[0006] Obtaining multi-modal data of each distributed energy of the virtual power plant, performing data cleaning and normalization processing on the multi-modal data to obtain processed distributed data, wherein the multi-modal data includes power output, meteorological data, and equipment operating status;

[0007] The processed distributed data is decomposed into three time scales of short-term, medium-term and long-term, each node of the virtual power plant and the relationship thereof are modeled as a graph structure, and GNN is used to capture the spatio-temporal dependence relationship to obtain a spatio-temporal feature vector of multiple time scales, wherein the nodes include a wind farm, a photovoltaic power station and an energy storage device, and the three time scales of short-term, medium-term and long-term are data of minute level to hour level, day level to week level and month level to year level respectively;

[0008] According to the feature vector, a self-supervised learning task is constructed for pre-training to extract deep features, and a contrast learning method is used to extract effective features between different time scales and modal data;

[0009] Based on the effective features, a normal data distribution model is generated by using a generative adversarial network, new normal sample data is obtained from the normal data distribution model, abnormal points deviating from the normal distribution are detected by comparing the new normal sample data with actual sample data, and abnormal data samples for enhancing model training are generated according to the abnormal points, and the abnormal data samples are summarized as a comprehensive data set;

[0010] According to the comprehensive data set, a multi-task learning framework is constructed, multi-task learning of abnormality detection, load prediction and fault diagnosis is performed, and a multi-task optimization model is obtained through shared features between tasks; based on the multi-task optimization model, an optimal abnormal response strategy is designed by using reinforcement learning, the abnormality detection model is optimized in real time through online learning and adaptive adjustment, and a final optimized abnormal response system is obtained, which is adapted to dynamic changes, and then the task of abnormality diagnosis of distributed data of the virtual power plant is completed through the abnormal response system.

[0011] Preferably, the processed distributed data is decomposed into three time scales of short-term, medium-term and long-term, wherein the short-term data is used to capture dynamic information of rapid changes in the power system, and the dynamic information includes instantaneous power fluctuation and instantaneous state change of equipment; the medium-term data is used to capture daily and periodic changes of the power system, and the daily and periodic changes include daily load fluctuation and weekend power consumption mode; and the long-term data is used to capture seasonal and annual trends of the power system, and the seasonal and annual trends include seasonal load change and annual energy consumption mode.

[0012] Preferably, each node of the virtual power plant and the relationship thereof are modeled as a graph structure, wherein the graph structure includes:

[0013] Each energy node, load node and substation node in the virtual power plant is modeled as a node in the graph;

[0014] The physical connection between nodes, power flow and geographical location relationship are modeled as edges in the graph, and the weight of the edge is set according to the power transmission capacity, distance and transmission efficiency, thereby forming a network graph containing all nodes and edges, wherein the network graph includes the topology and operating state of the virtual power plant.

[0015] Preferably, the GNN is used to capture the spatio-temporal dependence relationship, and a multi-time scale spatio-temporal feature vector is obtained, which includes:

[0016] Based on the three time scales and the graph structure, the graph neural network is used for spatio-temporal convolution operation, and the data of each node at different time scales is processed by graph convolution and time convolution;

[0017] Combined with time convolution, the dependence relationship of nodes in the time dimension is captured, and a graph feature vector containing spatio-temporal features is obtained;

[0018] The graph feature vector is processed by feature extraction and fusion, which includes the feature vector of each node, the feature fusion at different time scales to obtain the first fusion feature, and the integration of the feature vectors of each node at different time scales based on the first fusion feature to obtain the final multi-time scale spatio-temporal feature vector.

[0019] Preferably, according to the feature vector, a self-supervised learning task is constructed for pre-training to extract deep features, and a contrast learning method is used to extract effective features between different time scales and modal data, which includes:

[0020] According to the multi-time scale spatio-temporal feature vector, a self-supervised learning task is constructed, which includes using an automatic encoder of time scale and modal data for pre-training, training the encoder and decoder for each node feature vector, reconstructing the input data and minimizing the reconstruction error to obtain deep features;

[0021] Based on the deep features, a contrast learning method is used to compare the feature vectors from different time scales and modal data, and for each node, a twin network is constructed to calculate the similarity loss of positive and negative samples to obtain optimized deep features.

[0022] According to the optimized deep features, the feature vectors of each node at short-term, medium-term and long-term time scales are integrated, and the effective feature vectors of different time scales and modal data are integrated through feature fusion technology, and finally a multi-time scale feature representation for anomaly detection and prediction tasks is obtained.

[0023] Preferably, based on the effective features, a normal data distribution model is generated using a generative adversarial network, new normal sample data is analyzed from the normal data distribution model, abnormal points deviating from the normal distribution are detected by comparing the new normal sample data with actual samples, and abnormal data samples for enhancing model training are generated according to the abnormal points, and the abnormal data samples are summarized into a comprehensive data set, which includes:

[0024] The generator part of the generative adversarial network is trained according to the effective features, which includes that the generator receives random noise as input and generates new samples similar to the actual normal data distribution, and in the training process, the difference between the generated samples and the actual samples is minimized;

[0025] Based on the trained generator, new normal samples are sampled from the noise distribution, the generated new samples are compared with the actual normal samples, and abnormal points deviating from the normal distribution are detected using an anomaly detection algorithm;

[0026] According to the detected abnormal points, abnormal data samples are generated using the generator of the generative adversarial network, wherein the abnormal data samples are abnormal noise inputs, and the abnormal data samples are summarized into a comprehensive data set.

[0027] Preferably, according to the comprehensive data set, a multi-task learning framework is constructed for multi-task learning of anomaly detection, load prediction and fault diagnosis, and a multi-task optimization model is obtained through shared features between tasks, which includes:

[0028] Based on the comprehensive data set, a multi-task learning framework is designed to simultaneously train anomaly detection, load prediction and fault diagnosis tasks, wherein the framework is used to improve overall performance by sharing a bottom feature extractor and utilizing the correlation between different tasks;

[0029] The input comprehensive data set is feature-extracted using a network structure, and in the multi-task learning framework of the neural network, shared features are extracted through shared hidden layers or shared convolution kernels, wherein the network structure includes a convolutional neural network, a recurrent neural network or a transformer;

[0030] The multi-task learning of anomaly detection, load prediction and fault diagnosis is trained according to a loss function;

[0031] After multi-task learning training, a comprehensive multi-task optimization model is obtained based on shared features, wherein the multi-task optimization model is used to process anomaly detection, load prediction and fault diagnosis tasks at the same time.

[0032] Preferably, the multi-task optimization model is used to design an optimal abnormal response strategy based on reinforcement learning, to optimize the abnormal detection model in real time through online learning and adaptive adjustment, to obtain a final optimized abnormal response system that adapts to dynamic changes, and to complete the task of virtual power plant distributed data abnormal diagnosis through the abnormal response system, which includes:

[0033] The multi-task optimization model is learned according to the abnormal response strategy designed based on reinforcement learning, wherein in the reinforcement learning environment, the state s is defined as the current system state, the action a is defined as the response strategy, and the reward r is defined as the effect after the system response, the Q-learning algorithm is used to maximize the cumulative reward by optimizing the strategy, and the optimal abnormal response strategy is obtained.

[0034] Based on the optimal abnormal response strategy, the reinforcement learning model is continuously updated to adapt to the dynamic changes of the virtual power plant system through real-time online learning, and the deep deterministic policy gradient algorithm is used to optimize the strategy parameters in the real-time data stream, and the final updated reinforcement learning model is obtained.

[0035] According to the updated reinforcement learning model, adaptive adjustment and optimization are performed through an adaptive dynamic programming method, model parameters are adjusted in real time, and an optimized abnormal detection model is obtained.

[0036] The optimized abnormal detection model is used to monitor and respond to abnormal conditions in the virtual power plant distributed data in real time, and the abnormal diagnosis task is completed.

[0037] In a second aspect, the present application also provides a virtual power plant distributed data abnormal diagnosis device based on a deep neural network, which comprises:

[0038] A memory is configured to store a computer program.

[0039] A processor is configured to execute the computer program to implement the steps of the virtual power plant distributed data abnormal diagnosis method based on a deep neural network.

[0040] In a third aspect, the present application also provides a readable storage medium, wherein the readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the virtual power plant distributed data abnormal diagnosis method based on a deep neural network.

[0041] The present application has the following advantages:

[0042] The application solves the problem that the traditional method is difficult to process complex data according to the preprocessing and fusion of multi-modal data, ensures the high quality of data input, models each node of the virtual power plant and its relationship into a graph structure through multi-time scale spatio-temporal data modeling, captures the spatio-temporal dependence relationship by using a graph neural network (GNN), realizes the spatio-temporal data modeling of multiple time scales, and effectively extracts the short-term, medium-term and long-term spatio-temporal features, enhances the ability to capture complex abnormal patterns, and improves the accuracy of anomaly detection.

[0043] According to the self-supervised learning and feature extraction, the application constructs a self-supervised learning task (such as future data point prediction and data reconstruction), performs deep feature extraction, and extracts effective features between different time scales and modal data by using a contrast learning method. This process significantly improves the feature expression ability of the model and the adaptability to different data patterns; through a generative adversarial network (GAN), a normal data distribution model and high-quality abnormal data samples are generated, so that the model can generate and learn more abnormal samples, thereby enhancing the robustness and generalization ability of the model.

[0044] According to multi-task learning and transfer learning, the application constructs a multi-task learning framework, simultaneously performs related tasks such as anomaly detection, load prediction and fault diagnosis, improves the detection effect through shared features between tasks, the multi-task learning framework improves the data utilization efficiency and improves the comprehensive performance of the model; the application designs an optimal abnormal response strategy based on reinforcement learning, optimizes the anomaly detection model in real time through online learning and adaptive adjustment, adapts to the dynamically changing system environment, and finally obtains an optimized abnormal response system, which can quickly respond to and handle abnormal situations, and ensure the stable operation of the virtual power plant.

[0045] The application solves the limitations of traditional methods in processing complex data of virtual power plants by constructing a virtual power plant distributed data anomaly diagnosis method based on a deep neural network, realizes intelligent management of distributed energy resources, improves the accuracy and response speed of anomaly detection, and realizes rapid deployment and application of the model through multi-task learning and transfer learning, and significantly improves the operation efficiency and safety of the virtual power plant.

[0046] Other features and advantages of the application will be set forth in the following description of the application, and in part will become apparent to those skilled in the art from the description, or recognized by practicing the application as embodied in the written description and examples. The objects and other advantages of the application will be realized and attained by the structure particularly pointed out in the written description and claims, and the appended drawings. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some of the embodiments of the present application, and therefore should not be considered as limiting the scope. For those skilled in the art, other related drawings can also be obtained without creative labor.

[0048] Figure 1 The flow chart of the virtual power plant distributed data anomaly diagnosis method based on deep neural network described in the embodiments of the present application is shown.

[0049] Figure 2 The structure diagram of the virtual power plant distributed data anomaly diagnosis equipment based on deep neural network described in the embodiments of the present application is shown.

[0050] In the figure: 800, a marking receiving device; 801, a processor; 802, a memory; 803, a multimedia component; 804, an I / O interface; 805, a communication component. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solutions and advantages of the embodiments of the present application more clear, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, not all. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0052] It should be noted that: similar numbers and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the present application, the terms "first", "second" and the like are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0053] Embodiment 1:

[0054] The present embodiment provides a virtual power plant distributed data anomaly diagnosis method based on deep neural network.

[0055] Referring to Figure 1 , the present method includes steps S100, S200, S300, S400 and S500.

[0056] S100. Acquire multimodal data of each distributed energy source in the virtual power plant, perform data cleaning and normalization on the multimodal data to obtain the processed distributed data, including power output, meteorological data and equipment operating status.

[0057] Understandably, in this step, within the virtual power plant system, the operational data of distributed energy resources such as wind power, photovoltaics, and energy storage devices are highly heterogeneous and multimodal. These data types include, but are not limited to, power output data, meteorological data (such as wind speed, temperature, and irradiance), and equipment operational status data (such as voltage, current, frequency, temperature, and fault status). Acquiring this multimodal data is fundamental to anomaly diagnosis methods. Data collection includes power output data, meteorological data, and equipment operational status data. Power output data comes from energy management systems or distributed energy monitoring systems, while meteorological data includes wind speed, wind direction, temperature, humidity, and irradiance. This data can be obtained through weather stations, satellite remote sensing, and IoT sensors. Equipment operational status data involves equipment operating parameters such as voltage, current, frequency, temperature, and equipment operational status (e.g., whether a fault has occurred). This data originates from the equipment's monitoring system and sensor network.

[0058] Specifically, the data cleaning step aims to remove noise and outliers from the data. For missing values, interpolation, mean imputation, and other methods can be used. By detecting and correcting sensor malfunctions, data entry errors, and abnormal fluctuations, the integrity and accuracy of the data are ensured. Commonly used data cleaning techniques include mean imputation, median imputation, interpolation, and more machine learning methods such as anomaly detection models. Secondly, the purpose of normalization is to transform data of different dimensions and ranges to the same scale to facilitate subsequent modeling and analysis. Common normalization methods include min-max normalization (scaling the data to the [0,1] interval) and standardization (converting the data into a distribution with a mean of 0 and a standard deviation of 1).

[0059] Therefore, by cleaning and normalizing the multimodal data of distributed energy in virtual power plants, the quality and consistency of the data can be significantly improved, reducing the impact of noise and outliers on subsequent analysis. Cleaned data better reflects the actual operating status and environmental conditions of the equipment, while normalization ensures compatibility between different data types, facilitating model processing and feature extraction.

[0060] S200, decompose the processed distributed data into short-term, medium-term and long-term time scales, model each node and its relationship in the virtual power plant as a graph structure, and use GNN to capture the spatio-temporal dependence relationship to obtain a spatio-temporal feature vector of multiple time scales, wherein the nodes include wind farms, photovoltaic power stations and energy storage devices, and the short-term, medium-term and long-term time scales are data of minute to hour, day to week and month to year, respectively.

[0061] It can be understood that the step S200 includes S201, S202 and S203:

[0062] S201, the short-term data is used to capture the dynamic information of the power system which changes rapidly, wherein the dynamic information includes instantaneous power fluctuation and instantaneous state change of the equipment; the medium-term data is used to capture the daily and periodic changes of the power system, wherein the daily and periodic changes include daily load fluctuation and weekend power consumption mode; the long-term data is used to capture the seasonal and annual trends of the power system, wherein the seasonal and annual trends include seasonal load change and annual energy consumption mode.

[0063] It should be noted that the short-term data generally refers to minute-level or hour-level data, which is used to capture the dynamic information of the power system which changes rapidly, such as instantaneous power fluctuation and instantaneous state change of the equipment. For example, the output power of wind power and photovoltaic power generation will fluctuate significantly in a short time due to weather conditions, and instantaneous power fluctuation data can help identify these changes. Using short-term data, sliding window technology and fast Fourier transform (FFT) and other methods can be used for analysis to identify rapidly changing abnormal patterns. The medium-term data is generally day-level or week-level data, which is used to capture the daily and periodic changes of the power system, typical daily changes include daily load fluctuation (such as morning peak and evening peak) and periodic changes (such as weekend power consumption mode), and for medium-term data, time series analysis methods such as autoregressive moving average model (ARIMA) and periodic pattern recognition algorithm can be applied to predict and detect daily and periodic abnormalities. Long-term data refers to month-level, quarter-level or annual-level data, which is used to capture the seasonal and annual trends of the power system, for example, the electricity consumption in winter and summer may increase significantly due to heating and cooling demand, and the annual energy consumption pattern can reflect changes in overall economic activity and energy policy. Long-term data analysis can use trend analysis, seasonal decomposition methods (such as STL decomposition) and long short-term memory network (LSTM) and the like, which are used to identify long-term trends and seasonal changes in abnormalities.

[0064] S202, model each energy node, load node and substation node in the virtual power plant as a node in the graph.

[0065] S203, model the physical connections, power flows and geographical location relationships between nodes as edges in the graph, where the weights of the edges are set according to the power transmission capacity, distance and transmission efficiency, thereby forming a network graph containing all nodes and edges, wherein the network graph includes the topology and operating state of the virtual power plant, and the calculation formula is as follows:

[0066] G = (V, E)

[0067] In the formula, V is the node set, and E is the edge set.

[0068] It should be noted that in step S202, the distributed energy nodes include wind turbines, photovoltaic panels and energy storage devices, etc. The attributes of these nodes include the power output of the device, the device state and the geographical location, etc. The load node represents the power consumption equipment or power consumption area, and the attributes of these nodes include the load demand, power consumption mode and geographical location, etc. The transformer station node represents the key node of power transmission and distribution, and the attributes of these nodes include the transmission capacity, transformer station state and geographical location, etc.

[0069] In step S203, in the graph model, not only the nodes need to be represented, but also the connection relationship between the nodes. The physical connection, power flow and geographical location relationship between the nodes are represented by edges in the graph, and the weight of the edge reflects the strength and nature of the connection. Among them, the physical connection represents the direct connection between power equipment and facilities, such as cable, wire and transformer connection, and each edge represents the physical connection between two nodes. According to the actual physical layout of the virtual power plant, it is determined which nodes have direct connection, and these connections are represented as edges in the graph. The power flow represents the path and direction of power transmission in the system. The weight of the edge can reflect the size of the transmitted power, transmission loss and transmission efficiency, etc. The geographical location relationship reflects the geographical distribution and spatial relationship of the nodes. Nodes with closer distance may have stronger connection. Using geographic information system (GIS) data, the geographical distance between nodes is calculated, and the distance information is used as part of the weight of the edge, to ensure that the model can reflect the spatial dependence.

[0070] That is, by modeling the physical connection, power flow and geographical location relationship between nodes as edges in the graph, and setting the weight of the edge according to the power transmission capacity, distance and transmission efficiency, a network graph containing all nodes and edges is formed. This network graph not only reflects the topology of the virtual power plant, but also contains the information of the operating state. This structured representation lays the foundation for the application of graph neural networks, enabling the model to capture complex spatio-temporal dependencies and improve the accuracy and effectiveness of anomaly detection. Through this detailed and dynamic network graph modeling, the operating state of the virtual power plant is comprehensively monitored and analyzed, enhancing the stability and response capability of the system.

[0071] It can be understood that in this step S200, S204, S205 and S206 are also included:

[0072] In S204, based on the three time scales and the graph structure, a spatio-temporal convolution operation is performed on each node in different time scales using a graph neural network to perform graph convolution and time convolution processing on the data of each node in different time scales. The calculation formula of the spatio-temporal convolution is as follows:

[0073]

[0074] In the formula, is the feature vector of node i at the kth layer, N(i) is the neighbor node set of node i, c ij is the normalized coefficient between nodes i and j, W (k) and is the weight matrix of the kth layer, and σ is the activation function.

[0075] In S205, the time convolution is combined to capture the dependency relationship of the node in the time dimension to obtain a graph feature vector containing spatio-temporal features.

[0076] In S206, feature extraction and fusion processing are performed on the graph feature vector, including feature fusion of the feature vector of each node in different time scales to obtain a first fusion feature, and integration of the feature vectors of each node in different time scales based on the first fusion feature to obtain a final multi-time-scale spatio-temporal feature vector.

[0077] It should be noted that specifically, the graph convolution is used to capture the spatial dependency relationship between nodes, and the time convolution is used to capture the dependency relationship of the node in the time dimension. Common time convolution methods include one-dimensional convolutional neural network (1D-CNN) and gated recurrent unit (GRU). Therefore, by using the graph neural network to perform the spatio-temporal convolution operation, the dependency relationship of each node in the virtual power plant in the spatial and time dimensions can be captured at the same time. This method not only improves the modeling ability of the complex relationship between nodes, but also enhances the ability to capture the dynamic changes in time sequence. The graph convolution captures the spatial correlation between nodes, and the time convolution captures the change trend of the node in different time scales, so that the model can more accurately identify abnormal patterns.

[0078] Specifically, in step S205, the dependency relationship in the time dimension is crucial in the spatio-temporal convolution operation. Data of different time scales are processed through the time convolution layer to extract patterns and features in the time series. By applying time convolution (such as 1D-CNN or GRU), data of different time scales are convolved, and the time convolution layer can identify patterns and trends in the time series. The results of time convolution further enhance the spatio-temporal features of the nodes, and these feature vectors contain the spatial relationship of the nodes in the graph structure and the dynamic information in the time series. The generation of spatio-temporal feature vectors provides high-quality data input for subsequent feature extraction and fusion processing, improving the accuracy and robustness of anomaly detection.

[0079] Specifically, in step S206, the feature vectors of each node at different time scales are fused to obtain first fused features. The feature fusion method can use weighted average, splicing or self-attention mechanism (such as Transformer) and the like. Based on the first fused features, the feature vectors of each node at different time scales are integrated to form the final multi-time scale spatio-temporal feature vectors. The integration method can be realized by weighted aggregation, ensemble learning and the like.

[0080] In summary, this step improves the representation ability and generalization ability of the model, making the anomaly detection more accurate and reliable. The feature fused vector provides high-quality input data for subsequent anomaly detection, load prediction and fault diagnosis, further enhancing the detection effect and adaptability of the system.

[0081] S300, according to the feature vector, pre-training is performed by constructing a self-supervised learning task to extract deep features, and a contrast learning method is used to extract effective features between different time scales and modal data.

[0082] It can be understood that this step S300 includes S301, S302 and S303.

[0083] S301, according to the multi-time scale spatio-temporal feature vector, a self-supervised learning task is constructed, which includes pre-training using an automatic encoder of time scale and modal data, training the encoder and decoder for each node feature vector, reconstructing the input data and minimizing the reconstruction error to obtain deep features;

[0084] It should be noted that the input data of the multi-time scale space-time feature vector contains the space-time feature vectors extracted from different time scales, which have been preliminarily processed and fused through the previous steps; the process of training the autoencoder is to map the input space-time feature vector X to a low-dimensional hidden space Z by the encoder, train the autoencoder by minimizing the reconstruction error, and after training, the hidden space representation Z extracted by the encoder is the deep feature, which contains the high-order information of the input data. This method can better represent the internal structure and pattern of the data, thereby providing higher quality input for subsequent anomaly detection and prediction tasks. The application of self-supervised learning greatly reduces the dependence on manually labeled data, improves the training efficiency and generalization ability of the model.

[0085] S302, based on the deep feature, using the method of contrast learning, learning to compare the feature vectors from different time scales and modal data, for each node, construct a twin network, calculate the similarity loss of positive and negative samples, and obtain the optimized deep feature;

[0086] S303, according to the optimized deep feature, integrating the feature vectors of each node in the short-term, medium-term and long-term time scales, integrating the effective feature vectors of different time scales and modal data through feature fusion technology, and finally obtaining the multi-time scale feature representation for anomaly detection and prediction tasks.

[0087] It should be noted that the deep features obtained in step S301 are taken as the input of contrastive learning, and a siamese network, that is, a neural network composed of two shared parameters, is used to process the paired samples, wherein the siamese network is used to learn the similarity between the feature vectors, and the data of different time scales and modalities are paired to construct positive sample pairs (similar data pairs, such as different time scale data of the same node) and negative sample pairs (dissimilar data pairs, such as time scale data of different nodes). In step S302, the similarity loss calculation is the loss function of the contrastive learning, which can be calculated by using the contrastive loss. By minimizing the contrastive loss, the siamese network is trained, and the deep features are optimized, so that the distance between the positive sample pairs is smaller, and the distance between the negative sample pairs is larger. By using the method of contrastive learning, the deep features can be further optimized to have better discrimination ability between different time scales and modalities. In step S303, for each node, the optimized deep features of the node at short-term, medium-term and long-term time scales are collected, and these feature vectors are spliced together to form a comprehensive feature vector containing multi-time scale information. Among them, the feature fusion technology can use various feature fusion technologies, including weighted average, splicing, attention mechanism, etc., and the attention mechanism is especially suitable for capturing the correlation and importance between different time scales. Through feature fusion, the final multi-time scale feature representation is obtained, which integrates the effective information of different time scales and modal data.

[0088] Therefore, it can be understood that through the feature integration and fusion processing, the final feature representation containing multi-time scale information is obtained, and the correlation between different time scales and modal data can also be captured. The fused feature representation provides more rich and accurate input data for anomaly detection and prediction tasks, and improves the prediction accuracy and detection effect of the model. The application of feature fusion technology greatly enhances the comprehensive representation ability of data, making the anomaly diagnosis of the virtual power plant more efficient and reliable.

[0089] S400, based on the effective features, generating a normal data distribution model using a generative adversarial network, analyzing new normal sample data from the normal data distribution model, detecting abnormal points deviating from the normal distribution by comparing the new normal sample data with the actual sample, and generating abnormal data samples for enhancing model training according to the abnormal points, and integrating the abnormal data samples into a comprehensive data set.

[0090] It can be understood that in this step S400, S401, S402 and S403 are included.

[0091] S401, training the generator part of the generative adversarial network according to the effective features, wherein the generator receives random noise as input and generates new samples similar to the actual normal data distribution, and in the training process, the difference between the generated samples and the actual samples is minimized;

[0092] S402, based on the trained generator, sampling new normal samples from the noise distribution, comparing the generated new samples with the actual normal samples, and detecting abnormal points deviating from the normal distribution using an anomaly detection algorithm;

[0093] S403, generating abnormal data samples using the generator of the generative adversarial network according to the detected abnormal points, wherein the abnormal data samples are abnormal noise inputs, and simultaneously summarizing the abnormal data samples into a comprehensive data set.

[0094] It should be noted that the effective feature vectors obtained in the previous steps are used as the basis data, and these feature vectors are processed through multiple layers, containing data features of different time scales and modalities, and the generator is trained. The adversarial training mechanism of the generative adversarial network makes the generator continuously improve the quality of the generated samples, and the finally generated samples can effectively simulate the distribution characteristics of the normal data. In step S402, noise vectors are sampled from the standard normal distribution, and the noise vectors are mapped to new samples by the generator G. Normal samples obtained from actual data are used, and an anomaly detection algorithm (such as a method based on density estimation or a method based on machine learning) is used to compare the generated samples and the actual samples to detect abnormal points deviating from the normal distribution. When detecting abnormal points, the difference measure between the generated samples and the actual samples can be calculated, common methods include Euclidean distance, Mahalanobis distance, etc., and by setting a threshold ∈, samples whose difference exceeds the threshold are identified as abnormal points. By comparing the generated new samples with the actual normal samples, abnormal points deviating from the normal distribution can be effectively detected, and the new samples generated by the generative adversarial network have similar distribution characteristics to the actual samples, so they can provide a reliable benchmark for anomaly detection.

[0095] It can be understood that in step S403, according to the anomaly detection result in step S402, abnormal points deviating from the normal distribution are identified, and the abnormal points are used as noise inputs to generate abnormal data samples through the generator. Collect the generated abnormal data samples, summarize the generated abnormal data samples to form a comprehensive data set for model training, including normal data and abnormal data, which can effectively enrich the training data set, so that the model can be exposed to more abnormal situations during the training process, helping to improve the model's ability to identify abnormal situations, thereby enhancing the robustness and accuracy of the system.

[0096] S500, constructing a multi-task learning framework according to the comprehensive data set, simultaneously performing multi-task learning of anomaly detection, load prediction and fault diagnosis, obtaining a multi-task optimization model through shared features between tasks; based on the multi-task optimization model, designing an optimal abnormal response strategy by reinforcement learning, optimizing the anomaly detection model in real time through online learning and adaptive adjustment, obtaining a final optimized abnormal response system adapted to dynamic changes, and then completing the task of distributed data anomaly diagnosis of the virtual power plant through the abnormal response system.

[0097] It can be understood that the step S500 includes S501, S502, S503 and S504.

[0098] S501, based on the comprehensive data set, designing a multi-task learning framework to simultaneously train anomaly detection, load prediction and fault diagnosis tasks, wherein the framework is used to improve the overall performance by sharing the bottom feature extractor and utilizing the correlation between different tasks;

[0099] It should be noted that the comprehensive data set contains normal and abnormal samples generated in the previous steps, covers features of multiple data modalities and time scales, and constructs a neural network architecture with shared layers and task-specific branches. The shared layer is used to extract the bottom features, and each task-specific branch is optimized for a specific task. Among them, the initial several layers of the network are designed as shared layers, which extract general features by processing all samples in the comprehensive data set. These features include spatiotemporal features of different time scales, device operating states, etc. Above the shared layer, the network is divided into three branches, each branch focusing on a specific task (anomaly detection, load prediction, fault diagnosis), and these tasks have certain correlation, which can help improve the learning effect of each task. Therefore, by sharing the bottom feature extractor, each task can utilize the relevant information of each other to improve the efficiency and accuracy of feature extraction; the correlation between tasks can help each task's model learn complex features better, improve the overall performance and generalization ability.

[0100] S502, using a network structure to extract features from the input comprehensive data set, in the multi-task learning framework of the neural network, shared features are extracted through shared hidden layers or shared convolution kernels, wherein the network structure includes a convolutional neural network, a recurrent neural network or a transformer;

[0101] It should be noted that the convolutional neural network is suitable for extracting spatial features, especially for processing spatio-temporal data and image data, and can extract shared features in space through shared convolution kernels, while the recurrent neural network is suitable for processing sequence data, especially time series, and can capture long-term dependencies of time series through shared hidden layers. In the multi-task learning framework, the initial several layers of the network are designed as shared layers, which include convolutional layers, hidden layers or self-attention layers, and these layers extract shared features that can be used for multiple tasks by processing the integrated data set. Among them, feature extraction is the process of the network processing the integrated data set to extract feature representations, which include spatio-temporal features of nodes, power output features, device state features, etc. In this step, by using appropriate network structures, multi-modal features and multi-time scale features in the integrated data set can be effectively extracted, and the design of shared layers enables different tasks to share bottom-level features, improving the efficiency and accuracy of feature extraction.

[0102] S503, training the multi-task learning of anomaly detection, load prediction and fault diagnosis according to the loss function, and the calculation formula is as follows:

[0103]

[0104] In the formula, λ anomaly , λ load , λ fault is the task weight coefficient, is the total loss function.

[0105] S504, after multi-task learning training, based on shared features, an integrated multi-task optimization model is obtained, wherein the multi-task optimization model is used to process anomaly detection, load prediction and fault diagnosis tasks at the same time.

[0106] It should be noted that in the multi-task learning framework, the model simultaneously receives training data for anomaly detection, load prediction, and fault diagnosis. Through the backpropagation algorithm, the parameters of the model are optimized under the joint action of multiple tasks. During the training process, the model uses shared feature layers to extract general features. These shared features provide a unified foundation for multiple tasks, allowing each task to benefit from them. The trained model contains optimized parameters for multiple tasks. The structure design and training process of the model enable it to efficiently handle multiple tasks at the same time. Through the multi-task learning framework, the model can balance the needs of different tasks while optimizing multiple tasks simultaneously, improving overall performance. The optimization goal is not only the accuracy of a single task, but also the comprehensive performance of multiple tasks. Therefore, the comprehensive multi-task optimization model can handle multiple tasks simultaneously, avoiding redundant calculations of multiple separate models and improving the overall efficiency of the system. Multi-task learning can utilize the correlation between tasks to improve the performance of each task, and the shared feature layer ensures that different tasks promote each other during the training process, improving the overall performance of the model.

[0107] It can be understood that in this step S500, S505, S506, S507 and S508 are also included:

[0108] S505, the multi-task optimization model is learned according to the abnormal response strategy designed by reinforcement learning, wherein in the reinforcement learning environment, the state s is defined as the current system state, the action a is the response strategy, and the reward r is the effect after the system response. Using the Q-learning algorithm, the cumulative reward is maximized by optimizing the strategy to obtain the optimal abnormal response strategy;

[0109] S506, based on the optimal abnormal response strategy, the reinforcement learning model is continuously updated to adapt to the dynamic changes of the virtual power plant system through real-time online learning, and the deep deterministic policy gradient algorithm is used to optimize the strategy parameters in the real-time data stream to obtain the final updated reinforcement learning model;

[0110] S507, according to the updated reinforcement learning model, the adaptive dynamic programming method is used for adaptive adjustment and optimization, the model parameters are adjusted in real time, and the optimized abnormal detection model is obtained;

[0111] It should be noted that in step S505, the abnormal response strategy is optimized by the Q-learning algorithm, which can select the optimal action in different states, maximize the cumulative reward of the system, and improve the fault response and recovery ability of the system. Specifically, in step S506, the system continuously receives new data, updates the model parameters in real time, ensures that the model can quickly adapt to changes, and through the combination of real-time online learning and DDPG algorithm, the reinforcement learning model can adaptively adjust the strategy parameters, optimize the response ability of the system in real time, and ensure that the best performance is maintained in a dynamically changing environment. And through the adaptive dynamic programming method, the system can adjust and optimize the model parameters in real time, further improve the accuracy and response speed of abnormal detection, and ensure that the virtual power plant can quickly recover to normal operation under abnormal conditions.

[0112] S508, using the optimized abnormal detection model, real-time monitoring and responding to abnormal conditions in distributed data of virtual power plant, completing the abnormal diagnosis task, the calculation formula is as follows:

[0113] y diagnosis =M final (x input )

[0114] In the formula, x input is the input data, and y diagnosis is the diagnosis result.

[0115] S500, according to the comprehensive data set, a multi-task learning framework is constructed, and multi-task learning of abnormal detection, load prediction and fault diagnosis is carried out at the same time, through the shared features between tasks, a multi-task optimization model is obtained; based on the multi-task optimization model, the optimal abnormal response strategy is designed by using reinforcement learning, the abnormal detection model is optimized in real time through online learning and adaptive adjustment, and the final optimized abnormal response system adapting to dynamic changes is obtained, and then the task of abnormal diagnosis of distributed data of virtual power plant is completed through the abnormal response system.

[0116] Embodiment 2:

[0117] Corresponding to the above method embodiment, a virtual power plant distributed data abnormal diagnosis device based on a deep neural network is also provided in this embodiment. The virtual power plant distributed data abnormal diagnosis device based on a deep neural network described below can be mutually corresponding and referred to with the virtual power plant distributed data abnormal diagnosis method based on a deep neural network described above.

[0118] Figure 2 is a block diagram of a virtual power plant distributed data abnormal diagnosis device 800 based on a deep neural network according to an exemplary embodiment. As shown in Figure 2As shown, the deep neural network based virtual power plant distributed data anomaly diagnosis device 800 includes a processor 801 and a memory 802. The deep neural network based virtual power plant distributed data anomaly diagnosis device 800 also includes one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0119] The processor 801 is configured to control overall operations of the virtual power plant distributed data anomaly diagnosis device based on deep neural network 800 to complete all or part of the steps in the virtual power plant distributed data anomaly diagnosis method based on deep neural network. The memory 802 is configured to store various types of data to support the operation of the virtual power plant distributed data anomaly diagnosis device based on deep neural network 800, which can include, for example, instructions for any application or method operating on the virtual power plant distributed data anomaly diagnosis device based on deep neural network 800, and application-related data, such as contact data, sent and received messages, pictures, audio, video, and the like. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk. The multimedia component 803 can include a screen and an audio component. The screen can be a touch screen, for example, and the audio component is configured to output and / or input audio signals. For example, the audio component can include a microphone configured to receive external audio signals. The received audio signals can be further stored in the memory 802 or transmitted through the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, which can be a keyboard, mouse, or button, etc. These buttons can be virtual buttons or physical buttons. The communication component 805 is configured to perform wired or wireless communication between the virtual power plant distributed data anomaly diagnosis device based on deep neural network 800 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, near field communication (NFC), 2G, 3G or 4G, or a combination of one or more of them, so the corresponding communication component 805 can include a Wi-Fi module, a Bluetooth module or an NFC module.

[0120] In an example embodiment, the deep neural network-based virtual power plant distributed data anomaly diagnosis device 800 can be implemented by one or more Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), controller, microcontroller, microprocessor or other electronic elements for executing the deep neural network-based virtual power plant distributed data anomaly diagnosis method described above.

[0121] In another example embodiment, a computer readable storage medium including program instructions is also provided, which when executed by a processor, implement the steps of the deep neural network-based virtual power plant distributed data anomaly diagnosis method described above. For example, the computer readable storage medium can be the memory 802 described above including program instructions executable by the processor 801 of the deep neural network-based virtual power plant distributed data anomaly diagnosis device 800 to complete the deep neural network-based virtual power plant distributed data anomaly diagnosis method described above.

[0122] Corresponding to the method embodiments above, in this embodiment, a readable storage medium is also provided, which can be referred to the deep neural network-based virtual power plant distributed data anomaly diagnosis method described above.

[0123] The computer program stored on the readable storage medium implements the steps of the deep neural network-based virtual power plant distributed data anomaly diagnosis method of the method embodiments described above when executed by a processor.

[0124] The readable storage medium can be specifically a U disk, a mobile hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and various readable storage media that can store program codes.

[0125] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0126] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A virtual power plant distributed data anomaly diagnosis method based on a deep neural network, characterized in that, The method comprises the following steps: acquiring multi-modal data of each distributed energy source of a virtual power plant, performing data cleaning and normalization processing on the multi-modal data to obtain processed distributed data, wherein the multi-modal data comprises power output, meteorological data and equipment operating state; decomposing the processed distributed data into three time scales of short-term, medium-term and long-term, modeling each node and the relationship of the nodes of the virtual power plant as a graph structure, and capturing spatio-temporal dependency relationships by using GNN to obtain spatio-temporal feature vectors of multiple time scales, wherein the nodes comprise wind farms, photovoltaic power stations and energy storage devices, and the three time scales of short-term, medium-term and long-term are data of minute to hour level, day to week level and month to year level, respectively; based on the feature vectors, pre-training is performed by constructing a self-supervised learning task to extract deep features, and a contrast learning method is used to extract effective features between different time scales and modal data; based on the effective features, a normal data distribution model is generated by using a generative adversarial network, new normal sample data is obtained from the normal data distribution model, abnormal points deviating from the normal distribution are detected by comparing the new normal sample data with actual samples, and abnormal data samples for enhancing model training are generated according to the abnormal points, and the abnormal data samples are summarized as a comprehensive data set; a multi-task learning framework is constructed according to the comprehensive data set, and multi-task learning of anomaly detection, load prediction and fault diagnosis is performed, and a multi-task optimization model is obtained through the shared features between tasks; based on the multi-task optimization model, an optimal abnormal response strategy is designed by using reinforcement learning, the abnormal detection model is optimized in real time through online learning and adaptive adjustment, and a final optimized abnormal response system suitable for dynamic changes is obtained, and then the task of abnormal diagnosis of distributed data of the virtual power plant is completed through the abnormal response system.

2. The deep neural network-based virtual power plant distributed data anomaly diagnosis method according to claim 1, characterized in that, The processed distributed data is decomposed into three time scales of short-term, medium-term and long-term, which comprises: short-term data is used to capture dynamic information that changes rapidly in the power system, wherein the dynamic information comprises instantaneous power fluctuation and instantaneous state change of equipment; medium-term data is used to capture daily and periodic changes in the power system, wherein the daily and periodic changes comprise daily load fluctuation and weekend power consumption mode; long-term data is used to capture seasonal and annual trends in the power system, wherein the seasonal and annual trends comprise seasonal load change and annual energy consumption mode. 3.The virtual power plant distributed data anomaly diagnosis method based on deep neural network according to claim 1, wherein, The nodes and the relationship of the virtual power plant are modeled as a graph structure, which comprises: each energy node, load node and substation node in the virtual power plant is modeled as a node in the graph; the physical connection, power flow and geographical location relationship between the nodes are modeled as edges in the graph, wherein the weight of the edge is set according to the power transmission capacity, distance and transmission efficiency, and then a network graph containing all nodes and edges is formed, wherein the network graph comprises the topological structure and operating state of the virtual power plant, and the calculation formula is as follows: G=(V,E) In the formula, V is a node set, and E is an edge set.

4. The deep neural network-based virtual power plant distributed data anomaly diagnosis method according to claim 1, characterized in that, The spatio-temporal dependency relationships are captured by using GNN to obtain spatio-temporal feature vectors of multiple time scales, which comprises: Based on three time scales and graph structure, the graph neural network is used for spatio-temporal convolution operation, the data of each node at different time scales is processed by graph convolution and time convolution, and the calculation formula of spatio-temporal convolution is as follows: wherein, is the feature vector of node i at the kth layer, N(i) is the set of neighbor nodes of node i, c ij is the normalized coefficient between nodes i and j, W (k) and is the weight matrix of the kth layer, and σ is an activation function. Combined with time convolution, the dependence of nodes in the time dimension is captured, and a graph feature vector containing spatio-temporal features is obtained; The graph feature vector is processed by feature extraction and fusion, which includes feature fusion of the feature vector of each node at different time scales to obtain first fusion features, and integration of the feature vectors of each node at different time scales based on the first fusion features to obtain the final multi-time scale spatio-temporal feature vector.

5. The deep neural network-based virtual power plant distributed data anomaly diagnosis method according to claim 1, characterized in that, According to the feature vector, the deep feature is extracted by constructing a self-supervised learning task, and the effective features between different time scales and modal data are extracted by using the contrast learning method, which includes: According to the multi-time scale spatio-temporal feature vector, a self-supervised learning task is constructed, which includes using the automatic encoder of time scale and modal data for pre-training, training the encoder and decoder of the feature vector of each node, reconstructing the input data and minimizing the reconstruction error to obtain deep features; Based on the deep features, the contrast learning method is used to compare the feature vectors from different time scales and modal data, and for each node, a twin network is constructed to calculate the similarity loss of positive and negative samples to obtain optimized deep features; According to the optimized deep features, the feature vectors of each node at short-term, medium-term and long-term time scales are integrated, and the effective feature vectors of different time scales and modal data are integrated by feature fusion technology, and finally the multi-time scale feature representation for anomaly detection and prediction tasks is obtained.

6. The deep neural network-based virtual power plant distributed data anomaly diagnosis method according to claim 1, characterized in that, Based on the effective features, a normal data distribution model is generated by using a generative adversarial network, new normal sample data is analyzed from the normal data distribution model, and abnormal points deviating from the normal distribution are detected by comparing the new normal sample data with the actual sample data, and abnormal data samples are generated for model training according to the abnormal points, and the abnormal data samples are summarized as a comprehensive data set, which includes: According to the effective features, the generator part of the generative adversarial network is trained, which includes that the generator receives random noise as input and generates new samples similar to the actual normal data distribution, and in the training process, the difference between the generated samples and the actual samples is minimized; Based on the trained generator, new normal samples are sampled from the noise distribution, the generated new samples are compared with the actual normal samples, and abnormal points deviating from the normal distribution are detected by using an anomaly detection algorithm; According to the detected abnormal points, the generator of the generative adversarial network is used to generate abnormal data samples, and the abnormal data samples are abnormal noise inputs, and the abnormal data samples are summarized as a comprehensive data set.

7. The deep neural network-based virtual power plant distributed data anomaly diagnosis method according to claim 1, characterized in that, According to the comprehensive data set, a multi-task learning framework is constructed, and multi-task learning of anomaly detection, load prediction and fault diagnosis is performed, and a multi-task optimization model is obtained through the shared features between tasks, which includes: Based on the comprehensive data set, a multi-task learning framework is designed to train anomaly detection, load prediction and fault diagnosis tasks simultaneously, wherein the framework is used to improve overall performance by sharing underlying feature extractors and utilizing the correlation between different tasks; The network structure is used to extract features from the input comprehensive data set, and in the multi-task learning framework of the neural network, shared features are extracted through sharing hidden layers or sharing convolution kernels, wherein the network structure includes convolutional neural network, recurrent neural network or transformer; According to the loss function, the multi-task learning of anomaly detection, load prediction and fault diagnosis is trained, and the calculation formula is as follows: In the formula, λ anomaly , λ load , λ fault is a task weight coefficient, is a total loss function; After multi-task learning training, based on shared features, a comprehensive multi-task optimization model is obtained, wherein the multi-task optimization model is used to process anomaly detection, load prediction and fault diagnosis tasks at the same time. 8.The virtual power plant distributed data anomaly diagnosis method based on deep neural network according to claim 1, wherein, Based on the multi-task optimization model, the optimal abnormal response strategy is designed by using reinforcement learning, the abnormal detection model is optimized in real time through online learning and adaptive adjustment, the final optimized abnormal response system is obtained which adapts to dynamic changes, and then the task of virtual power plant distributed data anomaly diagnosis is completed through the abnormal response system, which includes: According to the abnormal response strategy designed by reinforcement learning, the multi-task optimization model is learned, wherein in the reinforcement learning environment, the state s is defined as the current system state, the action a is the response strategy, and the reward r is the effect after the system response, the Q-learning algorithm is used, the cumulative reward is maximized through the optimization strategy, and the optimal abnormal response strategy is obtained; Based on the optimal abnormal response strategy, the reinforcement learning model is updated in real time through online learning to adapt to the dynamic changes of the virtual power plant system, and the deep deterministic policy gradient algorithm is used to optimize the strategy parameters in the real-time data stream, and the final updated reinforcement learning model is obtained; According to the updated reinforcement learning model, adaptive dynamic programming method is used for adaptive adjustment and optimization, real-time feedback adjustment of model parameters is realized, and the optimized abnormal detection model is obtained; Using the optimized abnormal detection model, the abnormal situation in the distributed data of the virtual power plant is monitored and responded in real time, and the abnormal diagnosis task is completed, and the calculation formula is as follows: y diagnosis = M final (x input ) where x input is the input data, y diagnosis is the diagnosis result.

Images