Detection method for photovoltaic reverse power transmission abnormity and related equipment
By constructing the graph structure of the photovoltaic system and using spectral clustering and graph neural network models, the problem of high leakage detection rate in photovoltaic power transmission detection is solved, and efficient detection of photovoltaic power transmission is achieved.
Patent Information
- Application Number
- CN202510424333.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-07
- Publication Date
- 2025-08-01
AI Technical Summary
The detection method of photovoltaic power transmission in the prior art cannot capture the abnormal propagation characteristics under the topological constraints of the power grid, resulting in a high leakage detection rate.
The graph structure of a distributed photovoltaic system is constructed, the electrical distance is used as the edge weight between nodes, the low-dimensional feature matrix is extracted through spectral clustering, and the abnormal score of nodes is calculated using the graph neural network model, and the detection accuracy is improved by combining the irradiance-temperature model.
By retaining the global structural information of the power grid topology, the detection accuracy of photovoltaic reverse power transmission is improved and abnormal states can be effectively identified.
Smart Images
Figure CN120408244A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of photovoltaic energy storage, and particularly to a detection method for abnormal photovoltaic reverse power transmission and related equipment. Background Art
[0002] Photovoltaic reverse power transmission generally refers to the phenomenon in a photovoltaic power generation system that when the electric energy generated by a photovoltaic power station or a distributed photovoltaic power generation device still has a surplus after meeting the local load demand, this surplus electric energy will be reversely transmitted to the power grid, which is also known as "surplus power on-grid". Reverse power transmission may cause problems such as local power grid voltage increase, power grid frequency fluctuation, and change of power grid power flow, which are likely to affect the stability of the power grid. Therefore, it is necessary to detect and control the situation of photovoltaic reverse power transmission.
[0003] Currently, the general method directly clusters the photovoltaic output curve through K-means clustering to detect the phenomenon of photovoltaic reverse power transmission. However, this method cannot capture the abnormal propagation characteristics under the grid topology constraints, and has a high missed detection rate.
[0004] In view of this, a detection method for abnormal photovoltaic reverse power transmission and related equipment is needed. Summary of the Invention
[0005] Aiming at the problem of high missed detection rate of photovoltaic reverse power transmission in the prior art, the present invention provides a detection method for abnormal photovoltaic reverse power transmission and related equipment, which can improve the detection accuracy of photovoltaic reverse power transmission. The specific technical solutions are as follows:
[0006] In a first aspect, an embodiment of the present application provides a detection method for abnormal photovoltaic reverse power transmission, including:
[0007] Construct a graph structure of a distributed photovoltaic system, define the photovoltaic inverters in the distributed photovoltaic system as nodes in the graph structure, the electrical parameters and operation data of the photovoltaic inverters as the attributes corresponding to the nodes, and the electrical distance between the photovoltaic inverters as the edge weights corresponding to the nodes; perform spectral clustering on the graph structure to extract a low-dimensional feature matrix of the graph structure; input the low-dimensional feature matrix into a preset graph neural network (graph convolutional network, GCN) model to obtain a first anomaly score of the nodes output by the graph neural network model; and output the reverse power transmission anomaly level of the nodes based on the first anomaly score.
[0008] Preferably, the electrical parameter includes an output curve; before determining the reverse power transmission anomaly level of the node based on the first anomaly score, the method further includes: calculating the predicted output of the node through a preset irradiance-temperature model; calculating a second anomaly score of the node based on the output curve and the predicted output; the determining the reverse power transmission anomaly level of the node based on the first anomaly score includes: determining the reverse power transmission anomaly level based on the first anomaly score and the second anomaly score.
[0009] Preferably, the calculation formula for the edge weight is:
[0010]
[0011] where W ij is the edge weight between node i and node j, Z ij is the equivalent impedance between node i and node j, σ represents the bandwidth parameter, and the exp() function is the exponential function.
[0012] Preferably, the spectral clustering of the graph structure to extract the low-dimensional feature matrix of the graph structure includes: calculating the adjacency matrix and the degree matrix based on the graph structure, where the elements in the adjacency matrix are the edge weights between two different nodes; calculating the normalized Laplacian matrix based on the adjacency matrix and the degree matrix, and the corresponding calculation formula is:
[0013] L = D -1 / 2 (D - W)D -1 / 2 ;
[0014] where L is the normalized Laplacian matrix, W is the adjacency matrix; D is the degree matrix, and the element D ii in D is the sum of the edge weights of all edges connected to node i;
[0015] extracting the eigenvectors corresponding to the first k smallest eigenvalues in the normalized Laplacian matrix; constructing the low-dimensional feature matrix based on the eigenvectors corresponding to the first k smallest eigenvalues. Where k is an integer greater than 0.
[0016] Preferably, after extracting the first k smallest eigenvalues to construct the low-dimensional feature matrix, the method further includes: performing K-means clustering on the low-dimensional feature matrix to obtain k community structures in the graph structure; using the k community structures as the corresponding k training labels, and training the graph neural network model based on the training labels and the low-dimensional feature matrix.
[0017] Preferably, the inputting the low-dimensional feature matrix into a preset graph neural network model to obtain the first anomaly score of the node includes: obtaining the node embedding vector of the node through the graph neural network model based on the low-dimensional feature matrix; calculating the first anomaly score of the node based on the node embedding vector.
[0018] Preferably, the graph neural network model includes a first graph convolutional layer and a second graph convolutional layer; the calculation formula of the first graph convolutional layer is:
[0019]
[0020] wherein, H (1) represents the output of the first graph convolutional layer, ReLU() is a non-linear activation function, is the normalized adjacency matrix, U is the low-dimensional feature matrix, and W (0) is the first weight matrix;
[0021] The calculation formula of the second convolutional layer is:
[0022]
[0023] wherein, H (2) represents the node embedding vector, Softmax() is the activation function, and W (1) is the second weight matrix.
[0024] In a second aspect, an embodiment of the present application provides a detection system for abnormal photovoltaic reverse power transmission, which is applied to the method described in the first aspect, and includes:
[0025] A construction module, configured to construct a graph structure of a distributed photovoltaic system, define the photovoltaic inverters in the distributed photovoltaic system as nodes, and the electrical distance between the photovoltaic inverters as the edge weight corresponding to the nodes;
[0026] A dimensionality reduction module, configured to perform spectral clustering on the graph structure and extract a low-dimensional feature matrix of the graph structure;
[0027] A scoring module, configured to input the low-dimensional feature matrix into a preset graph neural network model to obtain a first abnormal score of the nodes output by the graph neural network model;
[0028] An output module, configured to output the abnormal reverse power transmission level of the nodes based on the first abnormal score.
[0029] In a third aspect, an embodiment of the present application provides a computing device, including: a memory for storing a program; a processor for loading the program to execute the method described in the first aspect.
[0030] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, where the computer-readable storage medium includes a stored program, and when the program runs, it controls the device where the computer-readable storage medium is located to execute the method described in the first aspect.
[0031] Compared with the prior art, the beneficial effects of the present invention are as follows: By constructing a graph structure of a distributed photovoltaic system, using the electrical distance as the edge weight between nodes, and then performing spectral clustering on the graph structure, the global structure information of the power grid topology can be extracted, mapping the high-dimensional graph structure to a low-dimensional feature space, and retaining key physical constraints; then, through a graph neural network, the abnormal score of the node can be effectively calculated based on the vectors in the low-dimensional feature space to distinguish the normal state and the abnormal state of the node. The present invention can improve the detection accuracy of photovoltaic reverse power transmission. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.
[0033] Figure 1 It is a schematic flowchart of a method for detecting photovoltaic reverse power transmission anomalies provided by an embodiment of the present application;
[0034] Figure 2 It is a schematic structural diagram of a system for detecting photovoltaic reverse power transmission anomalies provided by an embodiment of the present application;
[0035] Figure 3 It is a schematic structural diagram of a computing device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.
[0037] It should be understood that when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.
[0038] It should also be understood that the terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms.
[0039] It should be further understood that the term "and / or" used in the specification and appended claims of the present invention refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.
[0040] To solve the problem of high undetected rate of photovoltaic reverse power transmission in the prior art, the present invention provides a method for detecting photovoltaic reverse power transmission anomalies and related devices, which can improve the detection accuracy of photovoltaic reverse power transmission.
[0041] Please refer to Figure 1 , Figure 1 which is a schematic flow chart of a method for detecting photovoltaic reverse power transmission anomalies provided by an embodiment of the present application. This method can be applied to a computing device, such as Figure 1 shown, and the method includes the following steps:
[0042] Step 101, the computing device constructs a graph structure of the distributed photovoltaic system.
[0043] Among them, the photovoltaic system is a power generation system that converts solar energy into electrical energy, including photovoltaic modules, a support system, a busbar equipment, a photovoltaic inverter, electrical equipment, and monitoring equipment.
[0044] Among them, the computing device can define the photovoltaic inverter in the distributed photovoltaic system as a node, the electrical parameters and operating data of the photovoltaic inverter as the attributes corresponding to the node, and the electrical distance between the photovoltaic inverters as the edge weight corresponding to the nodes.
[0045] Among them, the computing device can communicate and connect with the photovoltaic inverters, monitoring equipment, or sensors deployed on various devices in the distributed photovoltaic system through wired or wireless means, and obtain the electrical parameters and operating data of each photovoltaic inverter in the distributed photovoltaic system in real time.
[0046] Preferably, the electrical parameters include an output curve, grid connection point voltage and current, and the operating data includes inverter efficiency, operating duration, and temperature.
[0047] After obtaining the above data, the computing device can perform preliminary processing on these data, including data cleaning, outlier detection, and compensation. Among them, data cleaning is mainly to remove obviously incorrect data points, such as values beyond the physically possible range; outlier detection can use the standard deviation method, that is, data points exceeding the range of plus or minus 3 times the standard deviation of the mean are marked as outliers and removed; data compensation is to perform interpolation processing on missing or abnormal data points, such as linear interpolation and spline interpolation.
[0048] Among them, the graph structure is a complex data structure composed of nodes (Vertices) and edges (Edges), which is used to represent objects and their relationships. Taking the electrical distance as the edge weight between nodes can quantify the electrical coupling strength between nodes, accurately characterize the topological structure of the power grid, and provide topological constraints for subsequent calculations.
[0049] Preferably, the calculation formula for the edge weight is:
[0050]
[0051] where W ij is the edge weight between node i and node j, Z ij is the equivalent impedance between node i and node j, σ represents the bandwidth parameter, and the exp() function is the exponential function.
[0052] Specifically, the electrical distance is usually a function of the equivalent impedance, and both physically reflect the electrical coupling strength between nodes. Therefore, the equivalent impedance is used here to calculate the electrical distance as the edge weight.
[0053] Specifically, the bandwidth parameter σ is used to determine the "width" of the function. It can control the attenuation speed of the weight and adjust the sparsity of the graph structure. The larger σ is, the slower the weight decays with the increase of distance, which means that even nodes with a relatively large distance may be assigned a higher weight, enhancing the global connectivity; the smaller σ is, the weight is only significant between nodes with a very short distance, enhancing the local connectivity and making the graph structure sparser.
[0054] Preferably, the value of σ is 0.1.
[0055] Among them, the computing device can calculate the equivalent impedance between corresponding nodes based on the electrical parameters of different nodes.
[0056] Step 102: The computing device performs spectral clustering on the graph structure and extracts the low-dimensional feature matrix of the graph structure.
[0057] Spectral clustering is a clustering algorithm based on graph theory. It realizes clustering by performing eigen-decomposition on the similarity matrix between data points and using the spectral properties of the matrix.
[0058] After constructing the graph structure, the computing device can perform spectral clustering on the graph structure. On the one hand, it can embed the physical topological constraints of the power grid into the input of the subsequent graph neural network model, that is, the low-dimensional feature matrix; on the other hand, it can also capture the global connection characteristics of the power grid through spectral decomposition, identify key nodes (such as hub nodes) and vulnerable nodes (such as end nodes).
[0059] In addition, the clustering results obtained by spectral clustering can obtain different community structures in the graph structure, and these community structures can be used as classification labels for nodes.
[0060] Preferably, the computing device can calculate the adjacency matrix and the degree matrix based on the graph structure, where the elements in the adjacency matrix are the edge weights between two different nodes; then calculate the normalized Laplacian matrix based on the adjacency matrix and the degree matrix, and the corresponding calculation formula is:
[0061] L = D -1 / 2 (D - W)D -1 / 2 ;
[0062] where L is the normalized Laplacian matrix, W is the adjacency matrix, and the elements in the adjacency matrix are the edge weights W between node i and node j ij ; D is the degree matrix, and the element D ii in D is the sum of the edge weights of all edges connected to node i. Then the computing device can extract the eigenvectors corresponding to the first k smallest eigenvalues in the normalized Laplacian matrix, and then construct the low-dimensional feature matrix based on the eigenvectors corresponding to the first k smallest eigenvalues.
[0063] where the first k smallest eigenvalues are: sorting the eigenvalues in the Laplacian matrix from smallest to largest, and the first k eigenvalues in the sorting are the first k smallest eigenvalues, and k is an integer greater than 0. Preferably, the calculation formula for the value of k can be:
[0064]
[0065] where N is the number of nodes, and Δ is the device type correction factor.
[0066] In spectral clustering, the eigenvector corresponding to the smallest eigenvalue of the normalized Laplacian matrix is a constant vector (all elements are equal), indicating the global connectivity of the graph; other eigenvalues correspond to the local structure information of the graph, and the corresponding eigenvectors can be used to divide the graph into k community structures. Therefore, spectral clustering can be considered as embedding the nodes of the graph structure into the low-dimensional space composed of the first k eigenvectors; in this low-dimensional space, since similar nodes will be closer geometrically, k-means clustering can effectively divide the community.
[0067] Therefore, before performing step 103, preferably, the computing device can perform K-means clustering on the low-dimensional feature matrix to obtain k community structures in the graph structure; using the k community structures as the corresponding k training labels, training the graph neural network model based on the training labels and the low-dimensional feature matrix.
[0068] By using the community structure as the training label and providing the initial classification information under the power grid topology constraint, the training times of the graph neural network can be greatly reduced, and the recognition ability of the model for unseen abnormal patterns can be enhanced.
[0069] Step 103: The computing device inputs the low-dimensional feature matrix into a preset graph convolutional network (GCN) model to obtain the first anomaly score of the nodes output by the graph neural network.
[0070] Among them, after obtaining the low-dimensional feature matrix that inherits the feature information of both node attributes and edge weights, the computing device can input the low-dimensional feature matrix into the preset GCN.
[0071] Among them, the GCN can be trained by the computing device, or can be trained by other devices and then transmitted to the computing device.
[0072] Preferably, the graph neural network model includes a first graph convolutional layer and a second graph convolutional layer; the calculation formula of the first graph convolutional layer is:
[0073]
[0074] Among them, H (1) represents the output of the first graph convolutional layer, ReLU() is a non-linear activation function, is the normalized adjacency matrix, U is the low-dimensional feature matrix, and W (0) is the first weight matrix.
[0075] Specifically, in practical applications, the adjacency matrix usually needs to be normalized to ensure numerical stability and the rationality of feature aggregation; the weight matrix is a learnable parameter that maps the input low-dimensional feature matrix to a new feature space.
[0076] The calculation logic of the first convolutional layer is: First, multiply the normalized adjacency matrix by the low-dimensional feature matrix to realize the aggregation of each node's neighbor features; then perform a linear transformation on the aggregated features through the weight matrix to extract more abstract features; then introduce non-linearity to enhance the model's expressive ability.
[0077] Optionally, the first weight matrix is randomly initialized and optimized through backpropagation.
[0078] The calculation formula of the second convolutional layer is:
[0079]
[0080] Among them, H (2) represents the node embedding vector, Softmax() is the activation function, and W (1)is the second weight matrix. Specifically, Softmax() is an activation function mainly used for multi-class classification tasks. Its core function is to convert the output of the neural network into a probability distribution, so that the prediction result of each node can be intuitively represented as the probability of belonging to different classes; the Softmax() function can convert any real number vector of the input into a probability distribution, ensuring that the value of each element is within the range of [0,1], and the sum of all elements is 1.
[0081] For example: the output node embedding vector H (2) is [0.2, 0.7, 0.1], indicating that the probability of this node belonging to the first class is 20%, the probability of belonging to the second class is 70%, and the probability of belonging to the third class is 10%.
[0082] Among them, the node embedding vector includes the GCN embedding vectors of each node.
[0083] Preferably, the computing device can obtain the node embedding vector of the node through the graph neural network model based on the low-dimensional feature matrix; based on the node embedding vector, calculate the first anomaly score of the node.
[0084] Among them, the calculation formula of the first anomaly score can be:
[0085]
[0086] Among them, h i is the GCN embedding vector of node i, u c is the class center, ||h i - μ c || is the distance between the GCN embedding vector of node i and the class center u c . This distance measures the degree of difference between node i and the class center in the feature space. The larger the distance, the greater the difference between node i and the class center; the smaller the distance, the closer node i is to the class center.
[0087] When the distance between node i and the class center is exactly the maximum value among the distances between all nodes and the class center, this ratio is 1; when node i coincides with the class center, that is, the distance is 0, this ratio is 0. Then subtract this ratio from 1 to get S, and the value range of S is also between 0 and 1. The closer the S value is to 1, the smaller the distance between node i and the class center, that is, the closer node i is to the class center, and the more similar the operating state of the inverter represented by node i is to the overall state represented by this class; on the contrary, the closer the S value is to 0, the larger the distance between node i and the class center, and the greater the difference between node i and the overall characteristics represented by the class, which means that there are abnormal situations of node i different from the whole.
[0088] Among them, the computing device can obtain the class center based on the clustering result of the spectral clustering method. Exemplarily, the computing device can calculate the mean value of the data points of each class as the class center.
[0089] Exemplarily, the computing device can obtain k class centers after performing K-means clustering on the low-dimensional feature matrix.
[0090] It can be understood that in the case of multiple class centers, the computing device can calculate the first anomaly score S of node i with respect to each class center struc respectively, and then take the maximum value among them as the first anomaly score.
[0091] It can be understood that from the graph structure to the GCN embedding vector, the main features retained include physical constraints and topological structure features of the power grid. Based on the GCN embedding vector, the anomalies of nodes in the topological structure can be measured, and nodes that deviate significantly from the normal mode (such as island operation) can be identified. Therefore, the first anomaly score can also be used as the structural anomaly score.
[0092] Step 104: The computing device outputs the reverse power transmission anomaly level of the node based on the first anomaly score.
[0093] Among them, the computing device can determine the reverse power transmission anomaly level of the node based on the first anomaly score and a preset score threshold, and output this information to an output device that can be perceived by maintenance personnel or management personnel.
[0094] In another possible scenario, the electrical parameter includes an output curve; the computing device can also calculate the predicted output of the node through a preset irradiance-temperature model; calculate the second anomaly score of the node based on the output curve and the predicted output; and then output the reverse power transmission anomaly level based on the first anomaly score and the second anomaly score.
[0095] Among them, the calculation formula of the second anomaly score can be:
[0096]
[0097] Among them, P actual is the actual output value of the node, P pred is the predicted output value of the node, and P rated is a preset reference output value. Based on the second anomaly score, electrical anomalies can be captured, and situations such as shadow occlusion and inverter failure causing abnormal output can be judged. Therefore, the second anomaly score can also be used as the electrical anomaly score.
[0098] Then, the computing device can perform weighted calculation on the first anomaly score and the second anomaly score to obtain a final score; and then output the reverse power transmission anomaly level based on the final score. Exemplarily, the weighting value of the first anomaly score is 0.6, and the weighting value of the second anomaly score is 0.4.
[0099] Preferably, the computing device can determine the action to be performed based on the final score and a preset score threshold.
[0100] Exemplarily, when the final score is greater than or equal to 0.8, the computing device can determine that the current reverse power transmission anomaly level is a severe anomaly and needs to stop running immediately for troubleshooting; when the final score is less than 0.8 and greater than 0.6, the computing device can determine that the current reverse power transmission anomaly level is a potential anomaly and needs to strengthen detection and record; when the final score is less than or equal to 0.6, the computing device can determine that the current reverse power transmission anomaly level is normal and only routine inspections need to be maintained.
[0101] In the embodiments of the present application, by constructing a graph structure of a distributed photovoltaic system, using the electrical distance as the edge weight between nodes, and then performing spectral clustering on the graph structure, the global structure information of the power grid topology can be extracted, mapping the high-dimensional graph structure to a low-dimensional feature space, and retaining key physical constraints; then, through a graph neural network, the anomaly score of the node can be effectively calculated based on the vectors in the low-dimensional feature space to distinguish the normal state and the abnormal state of the node. The present invention can improve the detection accuracy of photovoltaic reverse power transmission.
[0102] The method provided in the embodiments of the present application has been described above, and the system provided in the embodiments of the present application will be described below.
[0103] Please refer to Figure 2 , Figure 2 which is a schematic structural diagram of a detection system for photovoltaic reverse power transmission anomalies provided in the embodiments of the present application. As Figure 2 shown, the system 200 includes:
[0104] A construction module 201, configured to construct a graph structure of a distributed photovoltaic system, define the photovoltaic inverters in the distributed photovoltaic system as nodes, and use the electrical distance between the photovoltaic inverters as the edge weight corresponding to the nodes;
[0105] A dimensionality reduction module 202, configured to perform spectral clustering on the graph structure to extract a low-dimensional feature matrix of the graph structure;
[0106] A scoring module 203, configured to input the low-dimensional feature matrix into a preset graph neural network model to obtain a first anomaly score of the nodes output by the graph neural network model;
[0107] An output module 204, configured to output the reverse power transmission anomaly level of the nodes based on the first anomaly score.
[0108] Preferably, the electrical parameter includes an output curve; the scoring module 203 is further configured to calculate the predicted output of the node through a preset irradiance-temperature model; calculate a second anomaly score of the node based on the output curve and the predicted output; specifically, the scoring module 203 is configured to determine the reverse power transmission anomaly level based on the first anomaly score and the second anomaly score.
[0109] Preferably, the calculation formula of the edge weight is:
[0110]
[0111] where, W ij is the edge weight between node i and node j, Z ij is the equivalent impedance between node i and node j, σ represents the bandwidth parameter, and the exp() function is the exponential function.
[0112] Preferably, the dimensionality reduction module 202 is specifically configured to calculate the adjacency matrix and the degree matrix based on the graph structure, and the elements in the adjacency matrix are the edge weights between two different nodes; calculate the normalized Laplacian matrix based on the adjacency matrix and the degree matrix, and the corresponding calculation formula is:
[0113] L = D -1 / 2 (D - W)D -1 / 2
[0114] where, L is the normalized Laplacian matrix, W is the adjacency matrix; D is the degree matrix, and the element D ii in D is the sum of the edge weights of all edges connected to node i;
[0115] Extract the first k smallest eigenvalues from the normalized Laplacian matrix; construct the low-dimensional feature matrix based on the eigenvectors corresponding to the first k smallest eigenvalues.
[0116] Preferably, the system 200 further includes: a clustering module 205, configured to perform K-means clustering on the low-dimensional feature matrix to obtain k community structures in the graph structure; a training module 206, configured to use the k community structures as the corresponding k training labels, and train the graph neural network model based on the training labels and the low-dimensional feature matrix.
[0117] Preferably, the scoring module 203 is specifically configured to obtain the node embedding vector of the node through the graph neural network model based on the low-dimensional feature matrix; calculate the first anomaly score of the node based on the node embedding vector.
[0118] Preferably, the graph neural network model includes a first graph convolutional layer and a second graph convolutional layer; the calculation formula of the first graph convolutional layer is:
[0119]
[0120] Among them, H (1) represents the output of the first graph convolutional layer, and ReLU() is a non-linear activation function. is the normalized adjacency matrix, U is the low-dimensional feature matrix, and W (0) is the first weight matrix;
[0121] The calculation formula of the second convolutional layer is:
[0122]
[0123] Among them, H (2) represents the node embedding vector, Softmax() is the activation function, and W (1) is the second weight matrix.
[0124] The detection system for abnormal photovoltaic reverse power transmission of a transmission line provided by the embodiments of the present application can be understood by referring to the corresponding content in the foregoing method embodiment part, and will not be repeated here.
[0125] As Figure 3 shown, Figure 3 is a possible logical structure schematic diagram of a computing device provided by an embodiment of the present application. The computing device 300 includes: a processor 301, a communication interface 302, a memory 303, and a bus 304. The processor 301, the communication interface 302, and the memory 303 are interconnected through the bus 304. In the embodiments of the present application, the processor 301 is used to control and manage the actions of the computing device 300. For example, the processor 301 is used to execute Figure 1 the steps in the embodiment and / or other processes of the technologies described herein. The communication interface 302 is used to support the computing device 300 to communicate. The memory 303 is used to store the program code and data of the computing device 300.
[0126] Among them, the processor 301 can be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field-programmable gate array or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute various exemplary logic blocks, modules, and circuits described in connection with the disclosure of this application. The processor can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a digital signal processor and a microprocessor, and so on. The bus 304 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, Figure 3 only a thick line is used to represent it in Figure 3 , but it does not mean that there is only one bus or one type of bus.
[0127] In another embodiment of the present application, a computer-readable storage medium is further provided. The computer-readable storage medium includes instructions that, when run on a computer, cause the computer to execute the above Figure 1 method described in the embodiment.
[0128] Those of ordinary skill in the art can realize that the units of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the components of each example have been generally described according to their functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.
[0129] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0130] In several embodiments provided by the embodiments of the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of the devices or units can be in an electrical, mechanical, or other form.
[0131] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0132] In addition, in each embodiment of the present invention, the functional units can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.
[0133] If the function 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 this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: USB flash drives, read-only memories (ROMs), random access memories (RAMs), mobile hard disks, magnetic disks, or optical discs, and other media that can store program codes.
[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and the description of the present invention.
Claims
1. A detection method for abnormal photovoltaic reverse power transmission, characterized in that, Including: Constructing a graph structure of a distributed photovoltaic system, where the photovoltaic inverters in the distributed photovoltaic system are nodes in the graph structure, the electrical parameters and operating data of the photovoltaic inverters are attributes corresponding to the nodes, and the electrical distance between the photovoltaic inverters is the edge weight corresponding to the edges between the nodes; Performing spectral clustering on the graph structure to extract a low-dimensional feature matrix of the graph structure; Inputting the low-dimensional feature matrix into a preset graph neural network model to obtain a first anomaly score of the nodes output by the graph neural network model; Based on the first anomaly score, outputting the reverse power transmission anomaly level of the nodes.
2. The method according to claim 1, characterized in that, The electrical parameters include an output curve; before determining the reverse power transmission anomaly level of the nodes based on the first anomaly score, the method further includes: Calculating the predicted output of the nodes through a preset irradiance-temperature model; Calculating a second anomaly score of the nodes based on the output curve and the predicted output; The determining the reverse power transmission anomaly level of the nodes based on the first anomaly score includes: Determining the reverse power transmission anomaly level based on the first anomaly score and the second anomaly score.
3. The method according to claim 1 or 2, characterized in that, The calculation formula for the edge weight is: W ij = exp(-Z ij 2 / 2σ 2 ); Among them, W ij is the edge weight between node i and node j, Z ij is the equivalent impedance between node i and node j, σ represents the bandwidth parameter, and the exp() function is the exponential function.
4. The method according to claim 1 or 2, characterized in that, The performing spectral clustering on the graph structure to extract a low-dimensional feature matrix of the graph structure includes: Calculating an adjacency matrix and a degree matrix based on the graph structure, where the elements in the adjacency matrix are the edge weights between two different nodes; Calculating a normalized Laplacian matrix based on the adjacency matrix and the degree matrix, and the corresponding calculation formula is: L = D -1 / 2 (D - W)D -1 / 2 ; where L is the normalized Laplacian matrix, W is the adjacency matrix, D is the degree matrix, and the element D ii in D is the sum of the edge weights of all edges connected to node i; Extracting the eigenvectors corresponding to the first k smallest eigenvalues of the normalized Laplacian matrix, where k is an integer greater than 0; Based on the eigenvectors corresponding to the first k smallest eigenvalues, constructing the low-dimensional feature matrix.
5. The method according to claim 4, characterized in that After extracting the first k smallest eigenvalues to construct the low-dimensional feature matrix, the method further includes: Performing K-means clustering on the low-dimensional feature matrix to obtain k community structures in the graph structure; Using the k community structures as the corresponding k training labels, and training the graph neural network model based on the training labels and the low-dimensional feature matrix.
6. The method according to claim 1 or 2, characterized in that, The inputting the low-dimensional feature matrix into a preset graph neural network model to obtain a first anomaly score of the nodes includes: Based on the low-dimensional feature matrix, obtaining a node embedding vector of the nodes through the graph neural network model; Calculating a first anomaly score of the nodes based on the node embedding vector.
7. The method according to claim 1 or 2, characterized in that, The graph neural network model includes a first graph convolutional layer and a second graph convolutional layer; the calculation formula of the first graph convolutional layer is: H (1) = ReLU(AUW (0) ); Among them, H (1) represents the output of the first graph convolutional layer, ReLU() is a non-linear activation function, A is the normalized adjacency matrix, U is the low-dimensional feature matrix, and W (0) is the first weight matrix; The calculation formula of the second convolutional layer is: H (2) = Softmax(AH (1) W (1) )); Among them, H (2) represents the node embedding vector, Softmax() is the activation function, and W (1) is the second weight matrix.
8. A detection system for abnormal reverse power transmission of photovoltaic power, characterized in that, Applied to the method according to any one of claims 1-7, including: A construction module for constructing a graph structure of a distributed photovoltaic system, where the photovoltaic inverters in the distributed photovoltaic system are nodes in the graph structure, the electrical parameters and operating data of the photovoltaic inverters are attributes corresponding to the nodes, and the electrical distance between the photovoltaic inverters is the edge weight corresponding to the edges between the nodes; A dimensionality reduction module for performing spectral clustering on the graph structure and extracting a low-dimensional feature matrix of the graph structure; A scoring module for inputting the low-dimensional feature matrix into a preset graph neural network model to obtain a first anomaly score of the node output by the graph neural network model; An output module for outputting an abnormal power reverse transmission level of the node based on the first anomaly score.
9. A computing device, characterized in that, Comprising: A memory for storing programs; A processor for loading the program to execute the method according to any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes instructions that, when run on a computer, cause the computer to execute the method according to any one of claims 1-7.