Fault diagnosis method for multi-terminal DC transmission lines based on WaveNet

Abstract

The present invention discloses a multi-terminal direct current transmission line fault diagnosis method based on WaveNet. First, some voltage and current signals in the multi-terminal flexible direct current transmission system are collected and preprocessed; secondly, the sample data set is manually labeled and divided into a sample training set, a sample verification set, and a sample test set; then, the built WaveNet model is trained until no overfitting occurs; finally, the optimal analysis model is deployed in practice to obtain the multi-terminal flexible direct current transmission line fault diagnosis analysis results. The method of the present invention avoids the uncertainty factors caused by human factors in the prior art to a certain extent, and the diagnosis results do not depend on the physical model, are not affected by factors such as line parameters, load capacity, and system impedance, and the model has strong adaptability and high accuracy.

Description

Technical Field

[0001] The invention belongs to the technical field of power system analysis, and in particular relates to a multi-terminal direct current transmission line fault diagnosis method based on WaveNet. Background Art

[0002] Flexible DC transmission systems based on modular multilevel converters have the characteristics of bidirectional power flow and flexible control, and are gaining more and more attention. Multi-terminal DC transmission systems are one of the effective means to solve the large-scale transmission and consumption of renewable energy, and therefore have been widely used. However, multi-terminal flexible DC transmission systems have low damping and a fast current rise rate after a fault, which puts higher requirements on the fault diagnosis method of transmission lines.

[0003] Most of the existing research on multi-terminal flexible DC transmission line fault diagnosis methods are based on physical model-driven methods, that is, through fault mechanism model analysis, the fault voltage or current signal characteristic quantity is established, and the fault threshold is manually set to perform transmission line fault diagnosis. However, there will inevitably be unreasonable assumptions or simplifications through model mechanism analysis, and it is difficult to model and analyze the influence of factors such as line parameter changes, load capacity changes, and impedance parameter changes, which will inevitably lead to high fault diagnosis misjudgment and low accuracy. Therefore, in the face of increasingly complex power grid structures, the electrical characteristic quantities after the fault occurs will change significantly, making it difficult for existing methods to achieve accurate transmission line fault diagnosis.

[0004] The data-driven approach uses data statistics and data mining methods to discover the physical nature and operating laws of the power system as shown by the data, and to see through the nature and relationship of things through data. Since such methods do not require assumptions or simplifications, nor do they rely on mechanisms, they are based on data that reflects the true situation of the system. Therefore, the data-driven approach shows unique advantages in the face of complex large power grids with increasing uncertainty, multiple factors working together, and increasing interactivity with the outside world, and is suitable for multi-terminal flexible DC transmission line fault diagnosis. Summary of the invention

[0005] The purpose of the present invention is to provide a multi-terminal DC transmission line fault diagnosis method based on WaveNet, which solves the problems in the prior art that the diagnosis results are affected by artificial uncertainty factors and the model adaptability and accuracy are not high enough.

[0006] The technical solution adopted by the present invention is:

[0007] The multi-terminal DC transmission line fault diagnosis method based on WaveNet is specifically carried out in the following steps:

[0008] Step 1: Collect partial voltage and current signals in the multi-terminal flexible DC transmission system, including at least the positive voltage, negative voltage, positive current, and negative current on a DC transmission line, and a three-phase AC voltage and AC current on the AC side of a line connected to the DC transmission line;

[0009] Step 2: performing normalization preprocessing on the voltage data and current data collected in step 1 to obtain normalized preprocessed voltage data and current data;

[0010] Step 3: When the multi-terminal flexible DC transmission system operates under different working conditions, a sample data set is formed by simulation or by collecting actual system operation historical data;

[0011] Step 4: Manually label the sample data set and divide it into sample training set, sample verification set, and sample test set;

[0012] Step 5: Build the overall WaveNet model structure;

[0013] Step 6: Use the sample training set in step 4 to train the WaveNet model to obtain the trained WaveNet model;

[0014] Step 7: For the trained WaveNet model, use the sample validation set in step 4 to determine whether the model is overfitting; if overfitting occurs, retrain the WaveNet model according to step 6 until overfitting does not occur and the optimal WaveNet model is obtained;

[0015] Step 8: For the optimal WaveNet model in step 7, use the sample test set in step 4 to test the model;

[0016] Step 9: Input the voltage data and current data collected in real time in the multi-terminal flexible DC transmission system into the optimal WaveNet model obtained in step 7, and the model outputs the line number where the fault is located; wherein the voltage data and current data include at least the positive voltage, negative voltage, positive current, and negative current on a DC transmission line, and a three-phase AC voltage and AC current on the AC side of a line connected to the DC transmission line.

[0017] The present invention is also characterized in that:

[0018] In step 1, the partial voltage and current signal collection is specifically as follows: by combining partial voltage measuring devices and current measuring devices installed in the multi-terminal flexible direct current transmission system, partial voltage and partial current signals are collected in the multi-terminal flexible direct current transmission system respectively.

[0019] Step 2 is specifically as follows: the amplitudes of the voltage signal and the current signal collected in step 1 are sampled and calculated according to the following formula (1), and all are normalized according to the following formula (2) to obtain preprocessed data;

[0020]

[0021] Where N is the number of data points collected in each cycle, x 1 , x 2 , x 3 , ..., x N The data values ​​corresponding to each sampling data point in each cycle in turn;

[0022]

[0023] Among them, x * is the normalized data output, x is the original collected data, and x max is the maximum value of the input sample data, x min is the minimum value in the input sample data.

[0024] In step 3, different operating conditions include different line parameters, different load capacities, different system impedances, different fault types and different fault lines; wherein different fault types include single-phase short circuit, two-phase short circuit and three-phase short circuit; different fault lines include DC side lines and AC side lines.

[0025] Step 4 is specifically as follows: numbering the DC side lines and AC side lines in the multi-terminal flexible DC transmission system to be analyzed from 1 to m, and labeling the sample data set in step 3, wherein each sample data includes an input item and an output item, the input item is the voltage data and current data after normalization preprocessing in step 2; the output item is the line number of the working condition corresponding to the input item;

[0026] The labeled sample data set is divided into 70% for sample training set, 20% for sample verification set, and 10% for sample test set.

[0027] Step 5 specifically includes: the input part, the WaveNet structure part and the output part of the deep learning model based on WaveNet, wherein;

[0028] The input part only contains one input layer, which processes the input data obtained in step 4 into matrix data that can be efficiently calculated and batch processed;

[0029] The WaveNet structure part includes multiple WaveNet structures. Each WaveNet structure part first passes through the causal convolution layer, and then enters the void causal convolution layer to obtain longer-term perception, and then sends it to the gate activation function. The value of the activation function is then processed by 1D convolution with a convolution kernel of 1×1 for dimensionality reduction. Part of the result of the dimensionality reduction processing is sent to the skip connection, and the other part is added to the input of the WaveNet structure and sent to the next layer of the neural network;

[0030] The output part includes a Softmax layer to obtain the line number where the fault is located.

[0031] Step 6 is specifically as follows: the WaveNet overall model structure of step 5 is trained using the sample training set of step 4, the back propagation algorithm is used to update the parameters, the Adam optimizer is used for training, the loss function is the cross-entropy loss function, the input data is the input item of each sample data in step 4, and the output data is the output item of each sample data in step 4.

[0032] The beneficial effects of the multi-terminal DC transmission line fault diagnosis method based on WaveNet of the present invention are: the method can autonomously learn the characteristic information corresponding to the faults occurring on different transmission lines directly from the collected partial voltage and current signals, avoiding the process of manual fault feature extraction or manual determination of fault thresholds, reducing the uncertainty factors caused by human factors, and improving the accuracy of the diagnosis results;

[0033] The method of the present invention does not rely on physical models, is not affected by factors such as line parameters, load capacity, system impedance, etc., and has strong model adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a flowchart of a multi-terminal DC transmission line fault diagnosis method based on WaveNet;

[0035] Figure 2 A four-terminal DC transmission system topology in an embodiment of a multi-terminal DC transmission line fault diagnosis method based on WaveNet;

[0036] Figure 3 Figure 2. Training iteration update diagram of accuracy in the WaveNet-based multi-terminal DC transmission line fault diagnosis method. DETAILED DESCRIPTION

[0037] The WaveNet-based multi-terminal DC transmission line fault diagnosis method of the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0038] like Figure 1As shown in the figure, firstly, some voltage and current signals in the multi-terminal flexible DC transmission system are collected and preprocessed; secondly, the sample data set is manually labeled and divided into sample training set, sample verification set, and sample test set; then, the built WaveNet model is trained until there is no overfitting phenomenon; finally, the optimal analysis model is deployed in practice to obtain the fault diagnosis analysis results of the multi-terminal flexible DC transmission line.

[0039] The present invention is based on WaveNet multi-terminal DC transmission line fault diagnosis method, such as Figure 1 As shown, the specific implementation steps are as follows:

[0040] Step 1: Collect some voltage and current signals in the multi-terminal flexible DC transmission system. This includes at least one positive voltage, negative voltage, positive current, and negative current on a DC transmission line, and one three-phase AC voltage and AC current on the AC side of the DC transmission line. The specific process is as follows:

[0041] Combination Figure 2 The four-terminal flexible DC transmission system shown in the figure, Figure 2 This is the topology of the four-terminal flexible direct current transmission system in the embodiment. In this embodiment, the positive voltage, negative voltage, positive current, negative current on the J4 line, and the three-phase AC voltage and AC current on the AC side of the P1 converter station are collected, totaling 10 voltage and current data.

[0042] Step 2: Perform normalization preprocessing on the 10 voltage and current data collected in step 1 to obtain the normalized preprocessed voltage and current data; the specific process is:

[0043] The amplitudes of the 10 voltage and current data numbers collected in step 1 are all calculated as shown in formula (1), and all are normalized as shown in formula (2) to obtain preprocessed data;

[0044]

[0045] Where N is the number of data points collected in each cycle, x 1 , x 2 , x 3 , ..., x N The data values ​​corresponding to each sampling data point in each cycle are in turn.

[0046]

[0047] Among them, x * is the normalized data output, x is the original collected data, and x max is the maximum value of the input sample data, x minis the minimum value in the input sample data.

[0048] Step 3: When the system is running under different working conditions, a sample data set is formed through simulation or by collecting actual system operation historical data. The specific process is as follows:

[0049] Using PSCAD simulation software, Figure 2 In the four-terminal flexible DC transmission system shown in the figure, on each line (including the AC side of S1 converter station, the AC side of S2 converter station, the AC side of P1 converter station, the AC side of P2 converter station, the DC side of J1 line, the DC side of J2 line, the DC side of J3 line, and the DC side of J4 line), for six types of faults (including single-phase short circuit, two-phase short circuit, three-phase short circuit on the AC side, and positive pole grounding, negative pole grounding, and inter-pole short circuit on the DC side), simulation is performed when the line parameters, load capacity, and system impedance are changed respectively to form a sample data set.

[0050] Step 4: Manually label the sample data set and divide it into sample training set, sample verification set, and sample test set; the specific process is as follows:

[0051] right Figure 2 The 20 lines (DC side lines and AC side lines) in the four-terminal flexible DC transmission system are numbered 1 to 20. All sample data sets in step 3 are labeled. Each sample data contains an input item and an output item. The input item is the voltage and current amplitude after preprocessing in step 2; the output item is the line number 1 to 20 corresponding to the input item.

[0052] The entire labeled sample data set is divided into 70% for sample training set, 20% for sample verification set, and 10% for sample test set.

[0053] Step 5: Build the overall WaveNet model structure; the specific process is:

[0054] The deep learning model based on WaveNet includes the input part, WaveNet structure part, and output part.

[0055] The input part only includes one input layer, which processes the input data obtained in step 4 into matrix data that can be efficiently calculated and batch processed, and the matrix size is a three-dimensional matrix. The total amount of sampling data used in the present invention is 12,000 sample data, each sample data contains 500 time steps, and each time step contains 1 (number of acquisition nodes on the AC side) × 6 (three-phase voltage, three-phase current) + 1 (number of acquisition nodes on the DC side) × 4 (positive voltage, positive current, negative voltage, negative current) = 10-dimensional data. Thus, the voltage and current matrix data of the observation node are obtained, and the matrix is ​​in the form of [12000, 500, 10];

[0056] The WaveNet structure part includes multiple WaveNet structures. Each WaveNet structure part first passes through the causal convolution layer, then enters the void causal convolution layer to obtain a longer perception, and then sends it to the gate activation function. The value of the activation function is then processed by a one-dimensional convolution with a convolution kernel of 1×1. In the present invention, the number of output lines m analyzed is 20, so the number of 1×1 convolution kernels is 1×20=20. The result after the 1×1 one-dimensional convolution dimensionality reduction process is sent to the skip connection on the one hand, and the other part is added to the input of the WaveNet structure and sent to the next layer of the neural network. If {x 1 , x 2 , ..., x n} (n is the time step) is the voltage and current matrix data obtained after the input part is processed. After the WaveNet structure part, the historical dependency of the current input time series can be extracted {h b1 ,h b2 , ..., h bn}.

[0057] The output part includes a Softmax layer to obtain the line number where the fault is located.

[0058] Step 6: Use the sample training set formed in step 4 to train the model in step 5; each training traverses each data in the sample training set, and each traversal is called a generation. The model parameters need to be obtained after training for multiple generations. The specific process is: use the sample training set in step 4 to train the model in step 5, use the back propagation algorithm to update the parameters, use the Adam optimizer to train, and the loss function is the cross-entropy loss function. The input data is the input item of each sample described in step 4, and the output data is the corresponding output item described in step 4. Try to use different hyperparameters to adjust the model during training to obtain the model after training.

[0059] The number of convolutional network layers, the number of neurons, the type of activation function, and the learning rate can be adjusted.

[0060] like Figure 3 As shown, Figure 3 This is the accuracy change trend during the model training process. As the number of iterations increases, the model accuracy increases, and the final accuracy reaches 99.4%.

[0061] Step 7: Use the sample validation set formed in step 4 to determine whether the model trained in step 6 is overfitting; if overfitting occurs, retrain the model according to step 6 until overfitting is no longer present and the optimal model is obtained.

[0062] Step 8: Test the model in step 7 using the sample test set formed in step 4;

[0063] Step 9: input the voltage data and current data collected in real time in the multi-terminal flexible DC transmission system into the optimal WaveNet model obtained in step 7, and the model outputs the line number where the fault is located; wherein the voltage data and current data include at least a positive voltage, a negative voltage, a positive current, and a negative current on a DC transmission line, and a three-phase AC voltage and AC current on the AC side of a line connected to the DC transmission line;

[0064] The specific process is: use the collected positive voltage, negative voltage, positive current, negative current on the J4 line, and the real-time data of the three-phase AC voltage and AC current on the AC side of the P1 converter station, obtain the voltage and current amplitudes after preprocessing in step 2, and send them to the optimal model obtained in step 7. The output data is the line number 1 to 20 where the fault is located.

[0065] The invented WaveNet-based multi-terminal DC transmission line fault diagnosis method avoids the uncertainty factors caused by human factors in the prior art to a certain extent, and the diagnosis result does not depend on the physical model and is not affected by factors such as line parameters, load capacity, and system impedance. The model has strong adaptability and high accuracy.

Claims

1. A multi-terminal DC transmission line fault diagnosis method based on WaveNet, characterized in that: Follow the steps below: Step 1: Collect partial voltage and current signals in the multi-terminal flexible DC transmission system, including at least the positive voltage, negative voltage, positive current, and negative current on a DC transmission line, and a three-phase AC voltage and AC current on the AC side of a line connected to the DC transmission line; Step 2: performing normalization preprocessing on the voltage data and current data collected in step 1 to obtain normalized preprocessed voltage data and current data; Step 3: When the multi-terminal flexible DC transmission system operates under different working conditions, a sample data set is formed by simulation or by collecting actual system operation historical data; Step 4: Manually label the sample data set and divide it into sample training set, sample verification set, and sample test set; Step 5: Build the overall WaveNet model structure; Step 6: Use the sample training set in step 4 to train the WaveNet model to obtain the trained WaveNet model; Step 7: For the trained WaveNet model, use the sample validation set in step 4 to determine whether the model is overfitting; if overfitting occurs, retrain the WaveNet model according to step 6 until overfitting does not occur and the optimal WaveNet model is obtained; Step 8: For the optimal WaveNet model in step 7, use the sample test set in step 4 to test the model; Step 9: Input the voltage data and current data collected in real time in the multi-terminal flexible DC transmission system into the optimal WaveNet model obtained in step 7, and the model outputs the line number where the fault is located; wherein the voltage data and current data include at least the positive voltage, negative voltage, positive current, and negative current on a DC transmission line, and a three-phase AC voltage and AC current on the AC side of a line connected to the DC transmission line.

2. The WaveNet-based multi-terminal DC transmission line fault diagnosis method according to claim 1, characterized in that: In step 1, the partial voltage and current signal collection is specifically as follows: by combining partial voltage measuring devices and current measuring devices installed in the multi-terminal flexible direct current transmission system, partial voltage and partial current signals are collected in the multi-terminal flexible direct current transmission system respectively.

3. The WaveNet-based multi-terminal DC transmission line fault diagnosis method according to claim 1, characterized in that: Step 2 is specifically as follows: the amplitudes of the voltage data and current data collected in step 1 are sampled and calculated according to the following formula (1), and all are normalized according to the following formula (2) to obtain preprocessed data; Where N is the number of data points collected in each cycle, x1, x2, x3, ..., x N The data values ​​corresponding to each sampling data point in each cycle in turn; Among them, x * is the normalized data output, x is the original collected data, and x max is the maximum value of the input sample data, x min is the minimum value in the input sample data.

4. The WaveNet-based multi-terminal DC transmission line fault diagnosis method according to claim 1, characterized in that: In step 3, different operating conditions include different line parameters, different load capacities, different system impedances, different fault types and different fault lines; wherein different fault types include single-phase short circuit, two-phase short circuit and three-phase short circuit; different fault lines include DC side lines and AC side lines.

5. The WaveNet-based multi-terminal DC transmission line fault diagnosis method according to claim 1, characterized in that: Step 4 is specifically as follows: numbering the DC side lines and AC side lines in the multi-terminal flexible DC transmission system to be analyzed from 1 to m, and labeling the sample data set in step 3, wherein each sample data includes an input item and an output item, the input item is the voltage data and current data after normalization preprocessing in step 2; the output item is the line number of the working condition corresponding to the input item; The labeled sample data set is divided into 70% for sample training set, 20% for sample verification set, and 10% for sample test set.

6. The WaveNet-based multi-terminal DC transmission line fault diagnosis method according to claim 1, characterized in that: Step 5 specifically includes: the input part, the WaveNet structure part and the output part of the deep learning model based on WaveNet, wherein; The input part only contains one input layer, which processes the input data obtained in step 4 into matrix data that can be efficiently calculated and batch processed; The WaveNet structure part includes multiple WaveNet structures. Each WaveNet structure part first passes through the causal convolution layer, and then enters the void causal convolution layer to obtain longer-term perception, and then sends it to the gate activation function. The value of the activation function is then processed by 1D convolution with a convolution kernel of 1×1 for dimensionality reduction. Part of the result of the dimensionality reduction processing is sent to the skip connection, and the other part is added to the input of the WaveNet structure and sent to the next layer of the neural network; The output part includes a Softmax layer to obtain the line number where the fault is located.

7. The WaveNet-based multi-terminal DC transmission line fault diagnosis method according to claim 5, characterized in that: Step 6 is specifically as follows: the WaveNet overall model structure of step 5 is trained using the sample training set of step 4, the back propagation algorithm is used to update the parameters, the Adam optimizer is used for training, the loss function is the cross-entropy loss function, the input data is the input item of each sample data in step 4, and the output data is the output item of each sample data in step 4.

Images