Photovoltaic inverter fault positioning method and system based on graph data driving

Abstract

The invention discloses a photovoltaic inverter fault positioning method and system based on graph data driving, and the method comprises the steps: synchronously collecting a plurality of groups of historical data of the current time series of the DC sides of all grid-connected photovoltaic inverters in a target main network, and constructing a corresponding label set; discrete wavelet transform is carried out on each group of current time sequences, and wavelet energy entropy is calculated, so that a discrete wavelet energy entropy matrix set is obtained; constructing a graph structure from the discrete wavelet energy entropy matrix set by using an Euclidean-graph transformation mode, and obtaining a weighted adjacent matrix set of the graph; training the constructed space-time diagram convolutional network model based on the discrete wavelet energy entropy matrix set and the weighted adjacent matrix set; the trained model is used for fault positioning of the photovoltaic inverter. According to the invention, the photovoltaic inverter with a fault can be rapidly and accurately positioned, and operation and maintenance personnel can conveniently process the fault in time.

Description

Technical Field

[0001] The present invention relates to the technical field of power electronic fault location, and in particular to a photovoltaic inverter fault location method and system driven by graph data. Background Art

[0002] Photovoltaic inverters play a vital balancing role in the entire power generation system, and their safety and reliability have attracted much attention. However, in the actual operation of the system, key components such as power tubes, freewheeling diodes, and filter inductors in the inverter circuit may have open circuits or short circuits, which will have an adverse effect on the performance of the inverter and the overall operation of the energy Internet; once a fault alarm occurs, it may not only cause damage to the device and equipment, but also endanger personal safety. Through effective fault location methods, the fault location in the photovoltaic inverter can be discovered and located in time, thereby reducing system losses and maintaining stable operation of the equipment; and accurate fault location can help engineers quickly take corresponding protective measures to avoid more serious accidents; this is not only conducive to improving the reliability and safety of the system, but also can extend the service life of the equipment and reduce maintenance costs. More importantly, these measures help to ensure the stable operation of the energy Internet, ensure the efficient use of clean energy, and promote the development of renewable energy. Through in-depth research and development of fault location technology for photovoltaic inverters, a solid technical guarantee can be provided for the continuous optimization of photovoltaic power generation systems.

[0003] At present, there are two main methods for fault location of photovoltaic inverters driven by graph data: the first is a method based on system mathematical model, which is based on mathematical model and combined with modern control theory and optimization ideas for positioning; this method can provide high-precision fault location in theory, but in practical applications it faces problems such as high complexity, large amount of calculation, poor adaptability, and insufficient robustness, which limits its wide application. The second is a method based on system input and output signal processing, which detects faults by analyzing the input and output signals of the system; although this method simplifies the fault location process to a certain extent, it still has the disadvantages of being sensitive to noise, strong environmental dependence, difficult to adjust model parameters, high real-time requirements, and limited fault location accuracy. Summary of the invention

[0004] In order to solve the deficiencies in the prior art, the present invention provides a photovoltaic inverter fault location method and system based on graph data drive, which can quickly and accurately locate the faulty photovoltaic inverter, making it convenient for operation and maintenance personnel to handle the fault in time.

[0005] The present invention adopts the following technical solution.

[0006] In a first aspect, the present invention provides a photovoltaic inverter fault location method based on graph data driving, the method comprising:

[0007] Synchronously collect multiple sets of historical data of the current time series of the DC side of all grid-connected photovoltaic inverters in the target main grid;

[0008] Perform discrete wavelet transform on each group of current time series and calculate wavelet energy entropy to obtain a discrete wavelet energy entropy matrix set;

[0009] Constructing a graph structure from the discrete wavelet energy entropy matrix set by using a Euclidean-graph transformation method to obtain a weighted adjacency matrix set of the graph;

[0010] Based on the discrete wavelet energy entropy matrix set and the weighted adjacency matrix set, training the constructed spatiotemporal graph convolutional network model;

[0011] The real-time data of the current time series on the DC side of each photovoltaic inverter is obtained, and the discrete wavelet energy entropy matrix and the weighted adjacency matrix of the graph corresponding to the real-time data are input into the trained spatiotemporal graph convolutional network model to obtain the location of the faulty photovoltaic inverter.

[0012] Optionally, the multiple groups of historical data I of the current time series of the DC side of all photovoltaic inverters in the target main grid are expressed by formula (1):

[0013]

[0014] In the formula, represents the current time series of the i-th grid-connected photovoltaic inverter in the j-th group of data, N represents the number of grid-connected photovoltaic inverters in the target main grid, and M represents the number of groups of sampling time series.

[0015] Optionally, the step of performing discrete wavelet transform on each group of current time series and calculating wavelet energy entropy comprises:

[0016] Discrete wavelet decomposition is performed on each current time series to obtain the approximate coefficients and detail coefficients of discrete wavelet transform as shown in formula (3):

[0017]

[0018] In the formula, Represents the current time series The approximate coefficients obtained by decomposition at the u+1th level; Represents the current time series The detail coefficient obtained by decomposition of the u+1th layer; Represents the current time series The approximate coefficient obtained by decomposition at the uth layer in , v represents the translation parameter, where u=1,2,…,U, v=1,2,…,V, U represents the number of decomposition layers, and V represents the total translation amount; Represents the current time series The discrete value of the scaling function at scale u and position v; Represents the current time series The discrete value of the mother wavelet function in scale u and position v;

[0019] The relative discrete wavelet energy of each layer of detail coefficients in each current time series under different translation parameters is calculated using formula (4):

[0020]

[0021] In the formula, Represents the current time series The detail coefficients obtained by decomposition at the uth layer are represents the relative discrete wavelet energy under the translation parameter v, |·| 2 represents the square of the modulus;

[0022] The discrete wavelet energy entropy of each layer of detail coefficients in each current time series is calculated using formula (5):

[0023]

[0024] In the formula, Current time series The discrete wavelet energy entropy of the detail coefficients of the u-th layer in;

[0025] Based on the discrete wavelet energy entropy of each current time series, the discrete wavelet energy entropy matrix of each group is constructed as shown in formula (6):

[0026]

[0027] Where, X j Represents the discrete wavelet energy entropy matrix of the current time series in the jth group of data;

[0028] Based on the constructed discrete wavelet energy entropy matrices, a discrete wavelet energy entropy matrix set X is obtained: X = {X1, X2, ..., X j ,…,X M}.

[0029] Optionally, the step of constructing a graph structure from the discrete wavelet energy entropy matrix set by using the Euclidean-graph transformation method includes:

[0030] Each grid-connected photovoltaic inverter is regarded as a node in the graph and is denoted as The discrete wavelet energy entropy matrix X j The i-th row of data in is represented as the i-th node The discrete wavelet energy entropy is used to determine the neighbors of each node in the graph using formula (7), thereby obtaining the connection relationship in the graph structure:

[0031]

[0032] In the formula, represents the node of the i-th grid-connected photovoltaic inverter; Indicates the i-th node in the j-th group of data Nodes with connection relationships, Represents the i-th node in the j-th group of data The neighbor set of Represents the i-th node and the mth node Cosine similarity of discrete wavelet energy entropy, ε represents the set threshold;

[0033] Based on the connection relationship in the graph structure, the weight between two nodes with a connection relationship is calculated by the threshold Gaussian kernel weight function of formula (8):

[0034]

[0035] In the formula, Representation Node and The weight of the edge between them, β is the bandwidth variance of the Gaussian function, and exp represents the exponential operation;

[0036] According to the weights between the nodes, the weighted adjacency matrix of each group of data is constructed as shown below:

[0037]

[0038] In the formula, A j Represents the weighted adjacency matrix of the jth group of data;

[0039] Based on the constructed weighted adjacency matrices, a weighted adjacency matrix set A is obtained: A = {A1, A2, ..., A j ,…,A M}.

[0040] Optionally, the constructed spatiotemporal graph convolutional network model includes multiple spatiotemporal graph convolutional layers, one fully connected layer and one softmax layer;

[0041] Each layer of the spatiotemporal graph convolution layer includes a first temporal convolution layer, a spatial convolution layer and a second temporal convolution layer connected in sequence, and is used to extract temporal features and spatial features between input data.

[0042] Optionally, the training steps of the spatiotemporal graph convolutional network model include:

[0043] Step 1: Perform symmetric normalization and maximum and minimum normalization on the weighted adjacency matrix set A and the discrete wavelet energy entropy matrix set X respectively to obtain the training set and

[0044] Step 2: Define the current number of layers of the spatiotemporal graph convolutional layer as l, initialize l = 0; define the output feature matrix of the lth spatiotemporal graph convolutional layer as H (l) ,initialization

[0045] Step 3: Based on the training set H (l) Perform the first time convolution, which is expressed as follows:

[0046]

[0047] In the formula, represents the output feature matrix of the first temporal convolutional layer in the l+1th spatiotemporal graph convolutional layer, They represent the three convolution kernels of the same shape in the first temporal convolution layer in the l+1th spatiotemporal convolution layer; Respectively represent the first time convolution layer pair H (l) The output matrix after three different convolution operations: Relu(·), Sigmoid(·), Represent three different types of activation functions respectively;

[0048] Step 4: Use formula (10) to and the adjacency matrix Perform spatial convolution:

[0049]

[0050] In the formula, represents the output feature matrix of the spatial convolution layer in the l+1th spatiotemporal graph convolution layer, and Represent five different first learning parameters respectively; and They represent five different second learning parameters respectively, and ∥ represents feature concatenation operation; and They represent the four feature matrices after different feature transformations; g (l+1) Represents splicing features;

[0051] Step 5: Based on the output feature matrix of the spatial convolution layer, perform the second temporal convolution using formula (11):

[0052]

[0053] In the formula, H (l+1) represents the output feature matrix of the l+1th spatiotemporal graph convolutional layer, They represent the three convolution kernels of the same shape in the second temporal convolution layer in the l+1th spatiotemporal convolution layer; Represent the second time convolution layer pair The output matrix after three different convolution operations;

[0054] Step 6: Let l=l+1, and determine whether the current number of layers l is equal to the set number of layers L; if not, return to step 3; if not, execute step 7;

[0055] Step 7: Substitute the output feature matrix H of the Lth spatiotemporal graph convolutional layer (L) Input the fully connected layer, and the dimension of each node feature vector output is 2;

[0056] Step 8: Input the output matrix of the fully connected layer into the softmax layer to obtain the probability of each inverter failing and operating normally. Select the one with the highest probability as the predicted label of the node, recorded as

[0057] Step 9: Calculate the predicted label using the cross entropy loss function And the loss of the actual label Y, and perform gradient descent until the loss function converges and stops training to obtain the trained spatiotemporal graph convolutional network model.

[0058] Optionally, the expression of the cross entropy loss function is as follows:

[0059]

[0060] In the formula, Represents the predicted label of the i-th grid-connected PV inverter in the j-th group of data.

[0061] In a second aspect, the present invention provides a photovoltaic inverter fault location system driven by graph data, which runs the steps of any method described in the first aspect of the present invention, and the system includes:

[0062] A synchronous acquisition module, used to synchronously acquire multiple sets of historical data of the current time series of the DC side of all grid-connected photovoltaic inverters in the target main grid;

[0063] A discrete wavelet transform module is used to perform discrete wavelet transform on each group of current time series and calculate wavelet energy entropy to obtain a discrete wavelet energy entropy matrix set;

[0064] A graph structure construction module is used to construct a graph structure from the discrete wavelet energy entropy matrix set by using a Euclidean-graph transformation method to obtain a weighted adjacency matrix set of the graph;

[0065] A network model training module, used to train the constructed spatiotemporal graph convolutional network model based on the discrete wavelet energy entropy matrix set and the weighted adjacency matrix set;

[0066] The fault location module is used to obtain the real-time data of the current time series on the DC side of each photovoltaic inverter, and input the discrete wavelet energy entropy matrix and the weighted adjacency matrix of the graph corresponding to the real-time data into the trained spatiotemporal graph convolutional network model to obtain the location of the faulty photovoltaic inverter.

[0067] In a third aspect, the present invention provides a terminal, including a processor and a storage medium;

[0068] The storage medium is used to store instructions;

[0069] The processor is used to operate according to the instructions to execute the steps of any one of the methods described in the first aspect of the present invention.

[0070] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the methods described in the first aspect of the present invention.

[0071] The beneficial effects of the present invention are as follows:

[0072] 1. Traditional fault location methods mainly rely on mathematical models or input and output signal processing, while the present invention constructs a graph structure of the photovoltaic inverter system in a graph data-driven manner, which can more effectively capture the complex relationships and mutual influence characteristics between devices and improve the accuracy and robustness of fault location.

[0073] 2. The photovoltaic inverter fault location method provided by the present invention first performs discrete wavelet decomposition on the current time series and calculates the discrete wavelet energy entropy, which can effectively extract useful information from the signal and improve the accuracy of signal processing and analysis; then constructs a graph structure from the discrete wavelet entropy matrix set and obtains a weighted adjacency matrix, which fully captures the topological information in the data; finally, the weighted adjacency matrix set and the discrete wavelet energy entropy matrix set are input into the spatiotemporal graph convolutional network model to fully capture the time and space information in the input data, so as to improve the problem that the traditional graph neural network can only capture part of the spatial information in the data and result in weak feature extraction ability; the model can handle the complex nonlinear relationships in the system and has self-learning ability, and improves the accuracy of fault location by continuously updating and optimizing the model parameters; compared with traditional methods, it can better adapt to complex situations in practical applications.

[0074] 3. The present invention can respond quickly after a fault occurs and provide real-time fault location information, thereby helping engineers to take corresponding protective measures in a timely manner to avoid more serious accidents. This is not only conducive to improving the reliability and safety of the system, but also can extend the service life of the equipment and reduce maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 This is a flow chart of a photovoltaic inverter fault location method driven by graph data in the present invention;

[0076] Figure 2 It is a schematic diagram of the structure of the spatiotemporal graph convolutional network model in the present invention;

[0077] Figure 3 A schematic diagram of a training accuracy and training loss function curve of an experimental simulation performed based on the method of the present invention;

[0078] Figure 4 This is a structural principle block diagram of the photovoltaic inverter fault location system driven by graph data in the present invention. DETAILED DESCRIPTION

[0079] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in the present invention are only part of the embodiments of the present invention, not all of the embodiments. Based on the spirit of the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the scope of protection of the present invention.

[0080] Embodiment 1:

[0081] Reference Figure 1 The embodiment of the present invention provides a photovoltaic inverter fault location method based on graph data driving, which specifically includes the following steps:

[0082] Step S1: synchronously collect multiple groups of historical data of the current time series of the DC side of all grid-connected photovoltaic inverters in the target main grid;

[0083] In this embodiment, multiple groups of current time series I are shown in formula (1):

[0084]

[0085] In the formula, represents the current time series of the i-th grid-connected photovoltaic inverter in the j-th group of data, N represents the number of grid-connected photovoltaic inverters in the target main grid, and M represents the number of groups of sampling time series.

[0086] Furthermore, the actual labels of each group of current time series are constructed to obtain the label set Y, which is used to calculate the loss function in the subsequent training of the spatiotemporal graph convolutional network model. The label set Y is shown in formula (2):

[0087]

[0088] In the formula, It represents the actual label of the i-th grid-connected photovoltaic inverter in the j-th group of data, where the fault label is recorded as 0 and the normal label is recorded as 1.

[0089] Step S2: performing discrete wavelet transform on each group of current time series and calculating wavelet energy entropy to obtain a discrete wavelet energy entropy matrix set;

[0090] As an embodiment of the present invention, step S2 specifically includes:

[0091] S2.1: Perform discrete wavelet decomposition on each current time series to obtain the approximate coefficients and detail coefficients of discrete wavelet transform as shown in formula (3):

[0092]

[0093] In the formula, Represents the current time series I j i The approximate coefficients obtained by decomposition at the u+1th level; Represents the current time series I j i The detail coefficients obtained by decomposition at the u+1th layer, Represents the current time series The approximate coefficient obtained by decomposition at the uth layer in , v represents the translation parameter, where u=1,2,…,U, v=1,2,…,V, U represents the number of decomposition layers, and V represents the total translation amount; Represents the current time series The discrete value of the scaling function at scale u and position v; Represents the current time series I j i The discrete value of the mother wavelet function in scale u and position v;

[0094] S2.2. Use formula (4) to calculate the relative discrete wavelet energy of each layer of detail coefficients in each current time series under different translation parameters:

[0095]

[0096] In the formula, Represents the current time series The detail coefficients obtained by decomposition of the u-th layer, represents the relative discrete wavelet energy under the translation parameter v, |·| 2 represents the square of the modulus;

[0097] S2.3. Calculate the discrete wavelet energy entropy of each layer detail coefficient in each current time series using formula (5):

[0098]

[0099] In the formula, Current time series The discrete wavelet energy entropy of the detail coefficients of the u-th layer in;

[0100] S2.4. Based on the discrete wavelet energy entropy of each current time series, the discrete wavelet energy entropy matrix of each group is constructed as shown in formula (6):

[0101]

[0102] Where, X j Represents the discrete wavelet energy entropy matrix of the current time series in the jth group of data;

[0103] S2.5, based on the constructed discrete wavelet energy entropy matrices, obtain a discrete wavelet energy entropy matrix set X: X = {X1, X2, ..., X j ,…,X M}.

[0104] Step S3: constructing a graph structure from the discrete wavelet energy entropy matrix set by using the Euclidean-graph transformation method to obtain a weighted adjacency matrix set of the graph;

[0105] As an embodiment of the present invention, step S3 specifically includes:

[0106] S3.1. Take each grid-connected photovoltaic inverter as a node in the graph and record it as: The discrete wavelet energy entropy matrix X j The i-th row of data in is represented as the i-th node The discrete wavelet energy entropy is used to determine the neighbors of each node in the graph using formula (7), thereby obtaining the connection relationship in the graph structure:

[0107]

[0108] In the formula, represents the node of the i-th grid-connected photovoltaic inverter; Indicates the i-th node in the j-th group of data Nodes with connection relationships, Represents the i-th node in the j-th group of data The neighbor set of Represents the i-th node and the mth node Cosine similarity of discrete wavelet energy entropy, ε represents the set threshold;

[0109] It is further explained that the weighted adjacency matrix is ​​a matrix form representing the graph structure, where each element of the matrix represents the weight of the edge between the corresponding nodes in the graph. The discrete wavelet energy entropy matrix reflects the energy distribution of the signal at different frequency components, and these matrices can be used to capture the features of the signal at different scales. These features can be mapped to the graph structure through certain transformations (such as Euclidean-graph transformation), so that the information in these matrices can be represented by a graph. Specifically, through the Euclidean-graph transformation, the discrete wavelet energy entropy matrix can be mapped to a graph structure, where the elements in the matrix (such as energy entropy values) can be used as the weights of the edges between the nodes in the graph, thereby constructing a weighted adjacency matrix. This graph structure can effectively represent the local and global features of the signal.

[0110] S3.2. Based on the connection relationship in the graph structure, the weight between two nodes with a connection relationship is calculated by the threshold Gaussian kernel weight function of formula (8):

[0111]

[0112] In the formula, Representation Node and The weight of the edge between them, β is the bandwidth variance of the Gaussian function, and exp represents the exponential operation;

[0113] S3.3. According to the weights between the nodes, the weighted adjacency matrix of each group of data is constructed as shown below:

[0114]

[0115] In the formula, A j Represents the weighted adjacency matrix of the jth group of data;

[0116] S3.4, based on the constructed weighted adjacency matrices, obtain a weighted adjacency matrix set A: A = {A1, A2, ..., A j ,…,A M}.

[0117] Step S4: Based on the discrete wavelet energy entropy matrix set and the weighted adjacency matrix set, the constructed spatiotemporal graph convolutional network model is trained;

[0118] A preferred but non-limiting embodiment, referring to Figure 2As shown in FIG, the spatiotemporal graph convolutional network model constructed in this embodiment includes three spatiotemporal graph convolutional layers, one fully connected layer and one softmax layer; The three spatiotemporal graph convolution layers are used to extract the temporal and spatial features between the input data. Then, the output feature matrix is ​​converted into a one-dimensional vector through a fully connected layer and finally input into the softmax layer to obtain the predicted label.

[0119] Each of the spatiotemporal graph convolution layers includes a first temporal convolution layer, a spatial convolution layer, and a second temporal convolution layer connected in sequence; and the temporal convolution kernel size is 1×3. Figure 2 Where W1, W2, W3, and W4 represent weight matrices for multi-dimensional feature transformation of the feature matrix output by the temporal convolution layer, and W5 represents the weight matrix in the spatial convolution operation.

[0120] Specifically, the training steps of the spatiotemporal graph convolutional network model include:

[0121] Step 1: Perform symmetric normalization and maximum and minimum normalization on the weighted adjacency matrix set A and the discrete wavelet energy entropy matrix set X respectively to obtain the training set and

[0122] Step 2: Define the current number of layers of the spatiotemporal graph convolutional layer as l, initialize l = 0; define the output feature matrix of the lth spatiotemporal graph convolutional layer as H (l) ,initialization

[0123] Step 3: Based on the training set H (l) Perform the first time convolution, which is expressed as follows:

[0124]

[0125] In the formula, represents the output feature matrix of the first temporal convolutional layer in the l+1th spatiotemporal graph convolutional layer, They represent the three convolution kernels of the same shape in the first temporal convolution layer in the l+1th spatiotemporal convolution layer; Represents the first time convolution layer pair H (l) The output matrix after three different convolution operations: Relu(·), Sigmoid(·), Represent three different types of activation functions respectively;

[0126] Step 4: Use formula (10) to and the adjacency matrix Perform spatial convolution:

[0127]

[0128] In the formula, represents the output feature matrix of the spatial convolution layer in the l+1th spatiotemporal graph convolution layer, and Represent five different first learning parameters respectively; and They represent five different second learning parameters respectively, and ∥ represents feature concatenation operation; and They represent the four feature matrices after different feature transformations; g (l+1) Represents splicing features;

[0129] Step 5: Based on the output feature matrix of the spatial convolution layer, perform the second temporal convolution using formula (11):

[0130]

[0131] In the formula, H (l+1) represents the output feature matrix of the l+1th spatiotemporal graph convolutional layer, They represent the three convolution kernels of the same shape in the second temporal convolution layer in the l+1th spatiotemporal convolution layer; Represent the second time convolution layer pair The output matrix after three different convolution operations;

[0132] Step 6: Let l=l+1, and determine whether the current number of layers l is equal to the set number of layers L; if not, return to step 3; if not, execute step 7;

[0133] Step 7: Substitute the output feature matrix H of the Lth spatiotemporal graph convolutional layer (L) Input the fully connected layer, and the dimension of each node feature vector output is 2;

[0134] Step 8: Input the output matrix of the fully connected layer into the softmax layer to obtain the probability of each inverter failing and operating normally. Select the one with the highest probability as the predicted label of the node, recorded as

[0135] Step 9: Calculate the predicted label using the cross entropy loss function And the loss of the actual label Y, and perform gradient descent until the loss function converges and stops training to obtain the trained spatiotemporal graph convolutional network model.

[0136] Furthermore, the expression of the cross entropy loss function is as follows:

[0137]

[0138] In the formula, Represents the predicted label of the i-th grid-connected PV inverter in the j-th group of data.

[0139] Step S5: Obtain real-time data of the current time series on the DC side of each photovoltaic inverter, and input the discrete wavelet energy entropy matrix and the weighted adjacency matrix of the graph corresponding to the real-time data into the trained spatiotemporal graph convolutional network model to obtain the location of the faulty photovoltaic inverter.

[0140] In order to verify the effectiveness of the photovoltaic inverter fault location method based on graph data drive provided by the present invention, in the simulation experiment of this embodiment, a power grid model containing multiple photovoltaic inverters is built through simulink, and the failure of photovoltaic inverters at different positions is simulated. Multiple groups of current time series on the DC side of all photovoltaic inverters are collected, and corresponding training sets are constructed to train the spatiotemporal graph convolutional network model. After the spatiotemporal graph convolutional network model for photovoltaic inverter fault location is trained, the discrete wavelet energy entropy matrix corresponding to the current time series on the DC side of all photovoltaic inverters under the real-time operation of the power grid model and the weighted adjacency matrix of the graph are input into the trained model to test the performance of the model; refer to Figure 3 The training accuracy and training loss function curves of the experimental simulation based on the method of the present invention are shown in Fig. Figure 3 It can be seen from the experimental results that the average test accuracy of the method of the present invention is as high as 99.12%, and convergence can be achieved within only 80 iterations of training, which greatly reduces the amount of calculation and the complexity of calculation, and can quickly and accurately locate the faulty photovoltaic inverter.

[0141] The beneficial effects of the present invention are as follows:

[0142] 1. Traditional fault location methods mainly rely on mathematical models or input and output signal processing, while the present invention constructs a graph structure of the photovoltaic inverter system in a graph data-driven manner, which can more effectively capture the complex relationships and mutual influence characteristics between devices and improve the accuracy and robustness of fault location.

[0143] 2. The photovoltaic inverter fault location method provided by the present invention first performs discrete wavelet decomposition on the current time series and calculates the discrete wavelet energy entropy, which can effectively extract useful information from the signal and improve the accuracy of signal processing and analysis; then constructs a graph structure from the discrete wavelet entropy matrix set and obtains a weighted adjacency matrix, which fully captures the topological information in the data; finally, the weighted adjacency matrix set and the discrete wavelet energy entropy matrix set are input into the spatiotemporal graph convolutional network model to fully capture the time and space information in the input data, so as to improve the problem that the traditional graph neural network can only capture part of the spatial information in the data and result in weak feature extraction ability; the model can handle the complex nonlinear relationships in the system and has self-learning ability, and improves the accuracy of fault location by continuously updating and optimizing the model parameters; compared with traditional methods, it can better adapt to complex situations in practical applications.

[0144] 3. The present invention can respond quickly after a fault occurs and provide real-time fault location information, thereby helping engineers to take corresponding protective measures in a timely manner to avoid more serious accidents. This is not only conducive to improving the reliability and safety of the system, but also can extend the service life of the equipment and reduce maintenance costs.

[0145] Embodiment 2:

[0146] like Figure 4 As shown, the present invention provides a photovoltaic inverter fault location system driven by graph data, the system is used to implement the steps of the method in the above embodiment 1, and the system specifically includes:

[0147] A synchronous acquisition module, used to synchronously acquire multiple sets of historical data of the current time series of the DC side of all grid-connected photovoltaic inverters in the target main grid;

[0148] A discrete wavelet transform module is used to perform discrete wavelet transform on each group of current time series and calculate wavelet energy entropy to obtain a discrete wavelet energy entropy matrix set;

[0149] A graph structure construction module is used to construct a graph structure from the discrete wavelet energy entropy matrix set by using a Euclidean-graph transformation method to obtain a weighted adjacency matrix set of the graph;

[0150] A network model training module, used to train the constructed spatiotemporal graph convolutional network model based on the discrete wavelet energy entropy matrix set and the weighted adjacency matrix set;

[0151] The fault location module is used to obtain the real-time data of the current time series on the DC side of each photovoltaic inverter, and input the discrete wavelet energy entropy matrix and the weighted adjacency matrix of the graph corresponding to the real-time data into the trained spatiotemporal graph convolutional network model to obtain the location of the faulty photovoltaic inverter.

[0152] The photovoltaic inverter fault location system driven by graph data provided in the embodiment of the present invention and the photovoltaic inverter fault location method driven by graph data provided in Example 1 are based on the same technical concept, and can produce the beneficial effects as described in Example 1. For the contents not described in detail in this embodiment, please refer to Example 1.

[0153] Embodiment three:

[0154] A terminal provided by an embodiment of the present invention includes a processor and a storage medium;

[0155] The storage medium is used to store instructions;

[0156] The processor is used to operate according to the instruction to execute the steps of the method according to any one of the first embodiments.

[0157] Embodiment 4:

[0158] An embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the steps of any method described in Embodiment 1 are implemented.

[0159] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0160] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not to be interpreted as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0161] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0162] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0163] 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 above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A photovoltaic inverter fault location method based on graph data drive, characterized in that: Methods include: Synchronously collect multiple sets of historical data on the DC side current time series of all grid-connected photovoltaic inverters in the target main grid; Perform discrete wavelet transform on each group of current time series and calculate the wavelet energy entropy to obtain a discrete wavelet energy entropy matrix set; Constructing a graph structure from the discrete wavelet energy entropy matrix set by using a Euclidean-graph transformation to obtain a weighted adjacency matrix set of the graph; Training the constructed spatiotemporal graph convolutional network model based on the discrete wavelet energy entropy matrix set and the weighted adjacency matrix set; Real-time data of the current time series on the DC side of each photovoltaic inverter is obtained, and the discrete wavelet energy entropy matrix and the weighted adjacency matrix of the graph corresponding to the real-time data are input into the trained spatiotemporal graph convolutional network model to obtain the location of the faulty photovoltaic inverter.

2. The photovoltaic inverter fault location method based on graph data drive according to claim 1, characterized in that: Formula (1) is used to express the multiple sets of historical data I of the current time series of the DC side of all photovoltaic inverters in the target main grid that are synchronously collected: Where, represents the current time series of the i-th grid-connected PV inverter in the j-th group of data, N represents the number of grid-connected PV inverters in the target main grid, and M represents the number of groups of sampling time series.

3. The photovoltaic inverter fault location method based on graph data drive according to claim 1 or 2, characterized in that: The steps of performing discrete wavelet transform on each group of current time series and calculating wavelet energy entropy include: Perform discrete wavelet decomposition on each current time series to obtain the approximate coefficients and detail coefficients of discrete wavelet transform as shown in formula (3): Where, Represents the current time series The approximate coefficients obtained by decomposition at the u+1th layer; Represents the current time series The detail coefficients obtained by decomposition of the u+1th layer; Represents the current time series The approximate coefficient obtained by decomposition at the u-th layer in , v represents the translation parameter, where u=1,2,…,U, v=1,2,…,V, U represents the number of decomposition layers, and V represents the total translation amount; Represents the current time series The discrete value of the scaling function at scale u and position v; Represents the current time series The discrete value of the mother wavelet function in scale u and position v; The relative discrete wavelet energy of each layer detail coefficient in each current time series under different translation parameters is calculated using formula (4): Where, Represents the current time series The detail coefficients obtained by decomposition of the u-th layer, represents the relative discrete wavelet energy under the translation parameter v, |·| 2 represents the square of the modulus; The discrete wavelet energy entropy of each layer detail coefficient in each current time series is calculated using formula (5): Where, Current time series The discrete wavelet energy entropy of the detail coefficients of the u-th layer in; Based on the discrete wavelet energy entropy of each current time series, the discrete wavelet energy entropy matrix of each group is constructed, as shown in formula (6): Where, X j Represents the discrete wavelet energy entropy matrix of the current time series in the jth group of data; Based on the constructed discrete wavelet energy entropy matrices, a discrete wavelet energy entropy matrix set X is obtained: X={X1, X2,…, X j ,…,X M }.

4. The photovoltaic inverter fault location method based on graph data drive according to claim 3, characterized in that: The step of constructing a graph structure from the discrete wavelet energy entropy matrix set by using the Euclidean-graph transformation method includes: Each grid-connected photovoltaic inverter is regarded as a node in the graph and is denoted as The discrete wavelet energy entropy matrix X j The i-th row of data in is represented as the i-th node The discrete wavelet energy entropy is used to determine the neighbors of each node in the graph using formula (7), thereby obtaining the connection relationship in the graph structure: Where, represents the node of the i-th grid-connected photovoltaic inverter; Indicates the i-th node in the j-th group of data Nodes with connection relationships, Represents the i-th node in the j-th group of data The neighbor set of Represents the i-th node and the mth node Cosine similarity of discrete wavelet energy entropy, ε represents the set threshold; Based on the connection relationship in the graph structure, the weight between two nodes with a connection relationship is calculated using the threshold Gaussian kernel weight function of formula (8): Where, Representation node and The weight of the edge between them, β is the bandwidth variance of the Gaussian function, and exp represents the exponential operation; According to the weights between the nodes, the weighted adjacency matrix of each group of data is constructed, as shown in the following formula: Where A j Represents the weighted adjacency matrix of the jth group of data; Based on the constructed weighted adjacency matrices, a weighted adjacency matrix set A is obtained: A={A1, A2,…, A j ,…,A M }.

5. The photovoltaic inverter fault location method based on graph data drive according to claim 1, characterized in that: The constructed spatiotemporal graph convolutional network model includes multiple spatiotemporal graph convolution layers, one fully connected layer and one softmax layer; Each layer of the spatiotemporal graph convolution layer includes a first temporal convolution layer, a spatial convolution layer, and a second temporal convolution layer connected in sequence, and is used to extract temporal features and spatial features between input data.

6. The photovoltaic inverter fault location method based on graph data drive according to claim 5, characterized in that: The training steps of the spatiotemporal graph convolutional network model include: Step 1: Perform symmetric normalization and maximum and minimum normalization on the weighted adjacency matrix set A and the discrete wavelet energy entropy matrix set X respectively to obtain the training set and Step 2: Define the current number of layers of the spatiotemporal graph convolution layer as l, initialize l = 0; define the output feature matrix of the lth spatiotemporal graph convolution layer as H (l) ,initialization Step 3: Based on the obtained training set, (l) Perform the first time convolution, which is expressed as follows: Where, represents the output feature matrix of the first temporal convolution layer in the l+1th spatiotemporal graph convolution layer, They represent the three convolution kernels of the same shape in the first temporal convolution layer in the l+1th spatiotemporal convolution layer; Represents the first time convolution layer pair H (l) The output matrix after three different convolution operations: Relu(·), Sigmoid(·), Represents three different types of activation functions; Step 4: Use formula (10) to and the adjacency matrix Perform spatial convolution: Where, represents the output feature matrix of the spatial convolution layer in the l+1th spatiotemporal graph convolution layer, and Represent five different first learning parameters respectively; and Represent five different second learning parameters respectively, ∥ represents the feature splicing operation; and Respectively represent the four feature matrices after different feature transformations; g (l+1) Represents splicing features; Step 5: Based on the output feature matrix of the spatial convolution layer, perform the second temporal convolution using formula (11): Where H (l+1) represents the output feature matrix of the l+1th spatiotemporal graph convolutional layer, They represent the three convolution kernels of the same shape in the second temporal convolution layer in the l+1th spatiotemporal convolution layer; Represent the second time convolution layer pair The output matrix after three different convolution operations; Step 6: Let l=l+1, and determine whether the current number of layers l is equal to the set number of layers L; if not, return to step 3; if not, go to step 7; Step 7: Substitute the output feature matrix H of the Lth spatiotemporal graph convolutional layer (L) Input the fully connected layer, and the dimension of each node feature vector output is 2; Step 8: Input the output matrix of the fully connected layer into the softmax layer to obtain the probability of each inverter failing or operating normally. The one with the highest probability is selected as the predicted label of the node, which is recorded as Step 9: Calculate the predicted label using the cross entropy loss function And the loss of the actual label Y, and perform gradient descent until the loss function converges and stops training to obtain the trained spatiotemporal graph convolutional network model.

7. The photovoltaic inverter fault location method based on graph data drive according to claim 6, characterized in that: The expression of the cross entropy loss function is as follows: Where, Represents the predicted label of the i-th grid-connected PV inverter in the j-th group of data.

8. A photovoltaic inverter fault location system driven by graph data, running the photovoltaic inverter fault location method driven by graph data according to any one of claims 1 to 7, characterized in that: The system includes: A synchronous acquisition module is used to synchronously acquire multiple sets of historical data of the DC side current time series of all grid-connected photovoltaic inverters in the target main grid; A discrete wavelet transform module is used to perform discrete wavelet transform on each group of current time series and calculate the wavelet energy entropy to obtain a discrete wavelet energy entropy matrix set; A graph structure construction module is used to construct a graph structure from the discrete wavelet energy entropy matrix set by using a Euclidean-graph transformation method to obtain a weighted adjacency matrix set of the graph; A network model training module, used to train the constructed spatiotemporal graph convolutional network model based on the discrete wavelet energy entropy matrix set and the weighted adjacency matrix set; The fault location module is used to obtain the real-time data of the current time series on the DC side of each photovoltaic inverter, and input the discrete wavelet energy entropy matrix and the weighted adjacency matrix of the graph corresponding to the real-time data into the trained spatiotemporal graph convolutional network model to obtain the location of the faulty photovoltaic inverter.

9. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images