Active power distribution network fault section positioning method based on space-time diagram neural network

Abstract

The invention discloses an active power distribution network fault section positioning method based on a space-time diagram neural network, and belongs to the technical field of power grid fault positioning, and the method comprises the following steps: S1, building a multi-node active power distribution network model; s2, setting different fault parameters, and collecting fault data and graph data; s3, establishing a fault feature vector according to the collected fault data; s4, constructing a space-time diagram convolutional neural network model, and setting related hyper-parameters; and S5, dividing a total fault data set, training the space-time diagram convolutional neural network model, and finally outputting a fault section result. According to the active power distribution network fault section positioning method based on the space-time diagram neural network, power distribution network space topological structure information and fault recording can be comprehensively considered, the fault feature extraction capability can be improved, and meanwhile, the fault section positioning method based on the space-time diagram neural network can be used for positioning the fault section of the active power distribution network by utilizing valuable information contained in zero, negative and positive sequences in phase voltage and current. The accuracy of asymmetric fault positioning can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid fault location, and in particular to a method for locating a fault section of an active distribution network based on a spatiotemporal graph neural network. Background Art

[0002] The expansion of the scale of modern distribution systems and the increase in the number of load connections make distribution systems more prone to failures due to random events such as lightning strikes, insulation breakdown, and improper operation, which affect people's production and life, and even cause considerable economic losses due to unexpected power outages. At the same time, with the grid connection of a large number of distributed power sources, the distribution network cannot maintain the original topology, and the change in the topology will also cause the change of the flow direction and operation mode of the distribution network, making the flow direction uncertain, which will change the flow distribution of the entire power grid and cause the reverse flow distribution of the urban power grid. At the same time, even if the flow direction can remain unchanged, when a short-circuit fault occurs, the short-circuit current value will be affected to a certain extent, resulting in a significant change in the short-circuit current value flowing through. Through protection, this change trend has a large difference. Determinism makes it difficult to accurately predict the fault point. Faced with the problems brought by DG grid connection to distribution network fault diagnosis and location, the traditional distribution network protection method is still used to solve the multi-source and multi-terminal network architecture of distributed distribution network protection, resulting in many technical difficulties. Therefore, it is urgent to carry out research on new principles of distributed generation distribution network protection, especially to solve the key problems of fault detection and location. Summary of the invention

[0003] The purpose of the present invention is to provide a method for locating fault sections of an active distribution network based on a spatiotemporal graph neural network, which can comprehensively consider the spatial topological structure information and fault recording of the distribution network, improve the ability to extract fault features, and at the same time utilize the valuable information contained in the zero (Z), negative (N) and positive (P) sequences (Z, N, P) in the phase voltage and current to improve the accuracy of asymmetric fault location.

[0004] To achieve the above object, the present invention provides a method for locating a fault section of an active distribution network based on a spatiotemporal graph neural network, comprising the following steps:

[0005] S1. Establish a multi-node active distribution network model;

[0006] S2, set different fault parameters and collect fault data and graph data;

[0007] S3. Establish a fault feature vector based on the collected fault data;

[0008] S4. Build a spatiotemporal graph convolutional neural network model and set relevant hyperparameters;

[0009] S5. Divide the total fault data set, train the spatiotemporal graph convolutional neural network model, and finally output the fault section results.

[0010] Preferably, in step S1, a multi-node active distribution network model is established using PSCAD or SIMULINK software, the power supply voltage and frequency, as well as the circuit type, inductance, resistance and capacitance per kilometer of each section transmission line are set, and three-phase voltage and current measuring devices are installed at the nodes.

[0011] Preferably, in step S2, different fault types are set in each line section, sampling time and sampling frequency are set, and PMU is used to sample data within a certain period of time after the fault occurs to collect the fault three-phase voltage and current and phase angle after the fault occurs.

[0012] Preferably, in step S2, an adjacency matrix is ​​established according to the topology of the power grid to obtain graph data.

[0013] Preferably, in step S3, the zero sequence, negative sequence and positive sequence of each phase voltage and current are calculated according to the phase voltage difference, phase current difference and fault phase angle difference before and after the fault, and these quantities are aggregated into a fault characteristic vector.

[0014] Preferably, the feature vector of fault type classification is defined as:

[0015]

[0016] Among them, φ t is the fault vector, ΔV abc is the three-phase voltage difference, ΔI abc is the three-phase current difference, is the three-phase voltage phase angle difference, is the phase angle difference of the three-phase current;

[0017] Use the following formula to calculate the zero sequence, negative sequence, and positive sequence values:

[0018]

[0019] Where, χ represents voltage or current, α=1∠120°=-0.5+j0.866, α 2 =1∠240°=-0.5-j0.866; α0 represents zero sequence, α1 represents positive sequence, and α2 represents negative sequence;

[0020] Finally, an enhanced feature vector definition for fault location is defined as:

[0021]

[0022] Among them, V ZNP is the sequence voltage, I ZNPis each sequence current (including positive sequence, negative sequence and zero sequence).

[0023] Preferably, in step S4, the spatiotemporal graph convolutional neural network model includes the following parts: an input layer for receiving spatiotemporal graph data; a spatiotemporal convolutional layer for performing convolution operations on both time and space dimensions; an activation function for introducing nonlinearity; a pooling layer for reducing the dimension of spatiotemporal features; and a fully connected layer for integrating the features output by the convolutional layer to generate a final prediction result.

[0024] Preferably, the spatiotemporal convolution layer combines the graph convolution layer and the temporal convolution layer, and the graph convolution layer is expressed as: Among them, H (l) is the node feature matrix of the lth layer, W (l) is the weight matrix, σ is the activation function, Is an adjacency matrix; if the input is X∈R N×T×F , the temporal convolution is expressed as: X′=Conv1D(X), where Conv1D represents a one-dimensional convolution operation that processes features in the time dimension; combining the graph convolution layer and the temporal convolution layer operations, the spatiotemporal graph convolution neural network model is described as and H (t+1,τ) =Conv1D(H (t+1) ).

[0025] Preferably, in step S5, all fault feature vector samples are divided into a training set and a test set, and each data in the training set is marked with a corresponding fault section label for training the spatiotemporal graph convolutional neural network model so that it has the function of locating the fault section through the measured fault voltage and current data, and then the test set samples are input, and finally the spatiotemporal graph convolutional neural network model will output the fault section location result.

[0026] Therefore, the present invention adopts the above-mentioned active distribution network fault section location method based on space-time graph neural network, uses the time feature extraction and spatial feature aggregation capabilities of the space-time graph neural network, comprehensively considers the spatial topological structure information and fault recording of the distribution network, and improves the fault feature extraction capability. At the same time, it uses the valuable information contained in the zero (Z), negative (N) and positive (P) sequences (Z, N, P) in the phase voltage and current to improve the accuracy of asymmetric fault location.

[0027] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a step diagram of an embodiment of a method for locating a fault section of an active distribution network based on a spatiotemporal graph neural network of the present invention;

[0029] Figure 2It is an active distribution network model diagram of an embodiment of a method for locating a fault section of an active distribution network based on a spatiotemporal graph neural network of the present invention;

[0030] Figure 3 It is an algorithm flow chart of FDL of an embodiment of a method for locating a fault section of an active distribution network based on a spatiotemporal graph neural network of the present invention. DETAILED DESCRIPTION

[0031] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0032] Example

[0033] like Figure 1 As shown, the present invention provides a method for locating fault sections of an active distribution network based on a spatiotemporal graph neural network, comprising the following steps:

[0034] S1. Establish a multi-node active distribution network model.

[0035] Use PSCAD or SIMULINK software to establish a multi-node active distribution network model, such as Figure 2 As shown in Figure 1, the network contains 33 nodes, representing a typical radial structure in an overhead transmission line. Each pair of adjacent nodes corresponds to a fault section, and the distribution line power supply voltage is 12kV and the frequency is 50Hz. The transmission line is a resistor-inductor (RL) circuit type, with the resistor set to 1.5042 ohms and the inductor set to 1.3554 / 2 / π / 50H. Except for node 1, each node is equipped with a three-phase voltage and current monitoring device.

[0036] Set the power supply voltage and frequency, as well as the circuit type, inductance, resistance and capacitance per kilometer of each section of the transmission line, and install three-phase voltage and current measuring devices at the nodes.

[0037] S2. Set different fault parameters and collect fault data and graph data.

[0038] Set different fault types in each line section, set sampling time and sampling frequency, use PMU to sample data within a certain period of time after the fault occurs, and collect the fault three-phase voltage and current and phase angle after the fault occurs. According to the grid topology, establish an adjacency matrix and obtain graph data.

[0039] This embodiment uses transient metal faults to verify the proposed fault diagnosis model, including 10 types of faults: single-phase grounding fault, two-phase short circuit fault, two-phase short circuit grounding fault, three-phase short circuit fault, etc. The established 33-node active distribution network contains 32 line sections, so the fault location task can be regarded as a 32×10=320 classification task. During the simulation process, the total sampling time after the fault is 0.05s, the sampling frequency is 1kHz, and 0.02s of fault data are obtained from each fault sample. In this case, the proposed model can use any 0.02 seconds of fault data within 0.05 seconds after the fault occurs to determine the fault type and location.

[0040] S3. Establish a fault feature vector based on the collected fault data.

[0041] According to the phase voltage difference, phase current difference and fault phase angle difference before and after the fault, the zero sequence, negative sequence and positive sequence of each phase voltage and current are calculated, and these quantities are combined into a fault feature vector.

[0042] In this embodiment, the voltage and current differences before and during the fault contribute to the fault location based on the admittance difference. Therefore, the feature vector of the fault type classification is defined as:

[0043] φ t ={ΔV,ΔI}

[0044] Among them, φ t is a complex variable, and its real and imaginary parts will affect the fault type and location. Therefore, it is necessary to redefine the characteristic vector of fault type classification, and define the amplitude voltage difference, amplitude current difference and phase angle difference of each phase as:

[0045]

[0046] Among them, φ t is the fault vector, ΔV abc is the three-phase voltage difference, ΔI abc is the three-phase current difference, is the three-phase voltage phase angle difference, is the phase angle difference of the three-phase current;

[0047] In addition to the feature vector φ for fault type classification t In addition, the zero (Z), negative (N) and positive (P) sequences (Z, N, P) in the phase voltage and current contain valuable information that can improve the accuracy of fault location. The main reason is that the behavior of the asymmetric fault LG, LL and LLG in the zero (Z), negative (N) and positive (P) sequences (Z, N, P) in terms of special amplitude changes and phase angle rotation can provide us with additional knowledge about the fault location. The values ​​of the three sequences are calculated using the following formula:

[0048]

[0049] Where, χ represents voltage or current, α=1∠120°=-0.5+j0.866, α 2 =1∠240°=-0.5-j0.866; α0 represents zero sequence, α1 represents positive sequence, and α2 represents negative sequence;

[0050] Finally, an enhanced feature vector definition for fault location is defined as:

[0051]

[0052] Among them, V ZNP For I ZNP It is each sequence voltage and each sequence current (including positive sequence, negative sequence and zero sequence).

[0053] In summary, according to the design concept of the fault feature vector, this embodiment adopts a fault location FDL algorithm, such as Figure 3 By using this algorithm, the information of fault type and fault location can be obtained.

[0054] S4. Build the spatiotemporal graph convolutional neural network STGCN model and set relevant hyperparameters.

[0055] The spatiotemporal graph convolutional neural network STGCN model consists of the following parts: an input layer for receiving spatiotemporal graph data; a spatiotemporal convolutional layer for performing convolution operations on both the time and space dimensions; an activation function for introducing nonlinearity; a pooling layer for reducing the dimensionality of spatiotemporal features; and a fully connected layer for integrating the features output by the convolutional layer to generate the final prediction results.

[0056] The core of building the spatiotemporal graph convolutional neural network STGCN model is to combine graph convolution and temporal convolution. The graph convolution layer is used to extract spatial features. The graph convolution layer is expressed as: Among them, H (l) is the node feature matrix of the lth layer, W (l) is the weight matrix, σ is the activation function, Is the adjacency matrix; the temporal convolution layer is used to extract temporal features. If the input is X∈R N×T×F , the temporal convolution is expressed as: X′=Conv1D(X), where Conv1D represents a one-dimensional convolution operation that processes features in the time dimension. Combining the graph convolution layer and the temporal convolution layer operations, the spatiotemporal graph convolution neural network STGCN model can be described as and H (t+1,τ) =Conv1D(H (t+1) ).

[0057] S5. Divide the total fault data set, train the spatiotemporal graph convolutional neural network model, and finally output the fault section results.

[0058] All fault feature vector samples are divided into training sets and test sets. Each data in the training set is labeled with the corresponding fault section label to train the spatiotemporal graph convolutional neural network model so that it has the function of locating the fault section through the measured fault voltage and current data. Then the test set samples are input, and finally the spatiotemporal graph convolutional neural network model will output the fault section location result.

[0059] In this embodiment, the original fault data contains 50 time points. After downsampling, each fault sample contains 25 time points in a time period of 0.02s. Therefore, each original fault sample can obtain 25 sub-samples for training, so the total number of samples obtained is 32′10′25=8000, of which 70% of the samples are selected for model training, and the remaining data are used for model testing. The subsampling process of fault data is suitable for real-time fault diagnosis. Therefore, the spatiotemporal graph convolutional neural network STGCN model can be applied to real-time fault diagnosis. The fault state parameters used for training are shown in Table 1:

[0060] Table 1. Fault parameter setting table

[0061]

[0062] Therefore, the present invention adopts the above-mentioned active distribution network fault section location method based on spatiotemporal graph neural network, which can comprehensively consider the spatial topological structure information and fault recording of the distribution network, improve the fault feature extraction capability, and at the same time utilize the valuable information contained in the zero (Z), negative (N) and positive (P) sequences (Z, N, P) in the phase voltage and current to improve the accuracy of asymmetric fault location.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for locating fault sections in active distribution networks based on spatiotemporal graph neural networks, characterized in that: The following steps are involved: S1. Establish a multi-node active distribution network model; S2, set different fault parameters and collect fault data and graph data; S3. Establish a fault feature vector based on the collected fault data; S4. Build a spatiotemporal graph convolutional neural network model and set relevant hyperparameters; S5. Divide the total fault data set, train the spatiotemporal graph convolutional neural network model, and finally output the fault section results.

2. According to claim 1, a method for locating fault sections of active distribution networks based on spatiotemporal graph neural networks is characterized in that: In step S1, a multi-node active distribution network model is established using PSCAD or SIMULINK software, the power supply voltage and frequency, as well as the circuit type, inductance, resistance and capacitance per kilometer of each section of the transmission line are set, and three-phase voltage and current measurement devices are installed at the nodes.

3. The method for locating fault sections of active distribution networks based on spatiotemporal graph neural networks according to claim 1 is characterized in that: In step S2, different fault types are set in each line section, sampling time and sampling frequency are set, and the PMU is used to sample data within a certain period of time after the fault occurs to collect the fault three-phase voltage and current and phase angle after the fault occurs.

4. The method for locating fault sections of active distribution networks based on spatiotemporal graph neural networks according to claim 1 is characterized in that: In step S2, an adjacency matrix is ​​established according to the power grid topology structure to obtain graph data.

5. The method for locating fault sections of active distribution networks based on spatiotemporal graph neural networks according to claim 1 is characterized in that: In step S3, the zero sequence, negative sequence and positive sequence of each phase voltage and current are calculated according to the phase voltage difference, phase current difference and fault phase angle difference before and after the fault, and these quantities are combined into a fault feature vector.

6. The method for locating fault sections of active distribution networks based on spatiotemporal graph neural networks according to claim 5 is characterized in that: The feature vector for fault type classification is defined as: Among them, φ t is the fault vector, ΔV abc is the three-phase voltage difference, ΔI abc is the three-phase current difference, is the three-phase voltage phase angle difference, is the phase angle difference of the three-phase current; Use the following formula to calculate the zero sequence, negative sequence, and positive sequence values: Where, χ represents voltage or current, α=1∠120°=-0.5+j0.866, α 2 =1∠240°=-0.5-j0.866; α0 represents zero sequence, α1 represents positive sequence, and α2 represents negative sequence; Finally, an enhanced feature vector definition for fault location is defined as: Among them, V ZNP is the sequence voltage, I ZNP is the sequence current.

7. The method for locating fault sections of active distribution networks based on spatiotemporal graph neural networks according to claim 1 is characterized in that: In step S4, the spatiotemporal graph convolutional neural network model includes the following parts: an input layer for receiving spatiotemporal graph data; a spatiotemporal convolutional layer for performing convolution operations on both time and space dimensions; an activation function for introducing nonlinearity; and a pooling layer for reducing the dimensionality of spatiotemporal features. The fully connected layer is used to integrate the features output by the convolutional layer to generate the final prediction result.

8. The method for locating fault sections of active distribution networks based on spatiotemporal graph neural networks according to claim 7 is characterized in that: The spatiotemporal convolution layer combines the graph convolution layer and the temporal convolution layer. The graph convolution layer is expressed as: Among them, H (l) is the node feature matrix of the lth layer, W( l ) is the weight matrix, σ is the activation function, Is an adjacency matrix; if the input is X∈R N×T×F , the temporal convolution is expressed as: X'=Conv1D(X), where Conv1D represents a one-dimensional convolution operation to process features in the time dimension; combining the graph convolution layer and the temporal convolution layer operations, the spatiotemporal graph convolution neural network model is described as and H (t+1,τ) =Conv1D(H (t+1) ).

9. The method for locating fault sections of active distribution networks based on spatiotemporal graph neural networks according to claim 1 is characterized in that: In step S5, all fault feature vector samples are divided into training sets and test sets, and each data in the training set is labeled with a corresponding fault section label to train the spatiotemporal graph convolutional neural network model so that it has the function of locating the fault section through the measured fault voltage and current data. Then, the test set samples are input, and finally the spatiotemporal graph convolutional neural network model will output the fault section location result.

Images