Fault identification method and system for direct-current power distribution network

Abstract

The invention relates to the technical field of direct-current power distribution network fault identification, in particular to a direct-current power distribution network fault identification method and system. The method comprises the following steps: acquiring fault electric quantity sample data from a power distribution automation system in a direct-current power distribution network; secondly, performing time-frequency decomposition on the fault electric quantity sample data by adopting discrete wavelet packet transformation to obtain a time-frequency matrix, and converting the time-frequency matrix into a pixel matrix of a time-frequency spectrogram; then, a label corresponding to the fault type of the direct-current power distribution network is constructed; and finally, extracting identified fault key features by using the improved deep residual shrinkage network, and realizing fault identification and classification. Therefore, the typical fault problem of the direct-current power distribution network is considered, the direct-current power distribution network can be helped to rapidly identify the fault type and classify the fault type, and the safety of a power system is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of DC distribution network fault identification, and in particular to a DC distribution network fault identification method and system. Background Art

[0002] With the rapid development of distributed energy resources and the rapid growth of DC loads, the probability of DC system failures is rapidly increasing. To ensure the safe and stable operation of DC distribution networks, it is necessary to promptly identify system faults. Therefore, to improve fault identification capabilities, it is necessary to conduct in-depth research on a DC distribution network fault identification method and system based on an improved deep residual shrinkage network.

[0003] In the prior art, the paper "Fault Diagnosis of Active Distribution Networks Based on Parallel Fusion Deep Residual Contraction Networks" uses a fault diagnosis model based on a parallel fusion deep residual contraction network to process fault recording signals, achieving the goal of reliable and accurate fault location and identification. The paper "Fault Identification Method for Smart Distribution Networks Based on LOF and SVM" uses a distribution network protection algorithm based on local outlier factor (LOF) detection to detect faults in smart distribution networks and ultimately classify fault types. However, none of these prior art methods consider typical fault issues in DC distribution networks, making it difficult to effectively implement fault identification and classification. Summary of the Invention

[0004] The present invention provides a DC distribution network fault identification method and system to solve the problem that the methods in the prior art do not consider the typical fault problems of the DC distribution network and are difficult to effectively implement fault identification and classification.

[0005] In order to achieve the above object, the present invention is implemented through the following technical solutions:

[0006] In a first aspect, the present invention provides a method for identifying a DC distribution network fault, comprising:

[0007] S1: Obtain fault power sample data from the distribution automation system in the DC distribution network;

[0008] S2: performing time-frequency decomposition on the fault power sample data using discrete wavelet packet transform to obtain a time-frequency matrix, and converting the time-frequency matrix into a time-frequency spectrum;

[0009] S3: Inputting the time-frequency spectrum into a fault identification network to obtain a fault type of the DC distribution network and a label corresponding to each fault type;

[0010] S4: Use the improved deep residual shrinkage network to extract and identify the key fault features, calculate the probability of mapping the key fault features to the corresponding labels of the fault types of the DC distribution network, and obtain the final fault identification results based on the probability.

[0011] In a second aspect, the present application provides a DC distribution network fault identification system, comprising:

[0012] A fault information acquisition unit is used to obtain fault power sample data from a distribution automation system in a DC distribution network;

[0013] a fault data processing unit configured to perform time-frequency decomposition of the fault power sample data using discrete wavelet packet transform to obtain a time-frequency matrix, and to convert the time-frequency matrix into a time-frequency spectrum; and further configured to input the time-frequency spectrum into a fault identification network to obtain a fault type of the DC distribution network and a label corresponding to each fault type;

[0014] The fault identification unit is used to extract and identify key fault features using an improved deep residual shrinkage network, calculate the probability of mapping the key fault features to the labels corresponding to the fault types of the DC distribution network, and obtain the final fault identification results based on the probability.

[0015] In a third aspect, the present application provides a DC distribution network fault identification system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method described in the first aspect when executing the computer program.

[0016] Beneficial effects:

[0017] The DC distribution network fault identification method provided by the present invention first obtains fault power sample data from the distribution automation system in the DC distribution network. Secondly, the discrete wavelet packet transform is used to perform time-frequency decomposition on the fault power sample data to obtain a time-frequency matrix, and the time-frequency matrix is ​​converted into a pixel matrix of a time-frequency spectrum. Then, labels corresponding to the DC distribution network fault types are constructed. Finally, the improved deep residual shrinkage network is used to extract the key features of the identified faults and realize fault identification and classification. In this way, the typical fault problems of the DC distribution network are taken into consideration, which can help the DC distribution network to quickly identify and classify fault types, effectively improving the safety of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 The figure is a flow chart of a DC distribution network fault identification method according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0019] The following is a clear and complete description of the technical solutions of the present invention. It should be understood that the embodiments described are only a portion of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are intended to fall within the scope of protection of the present invention.

[0020] Unless otherwise defined, the technical or scientific terms used in the present invention shall have the usual meanings understood by persons of ordinary skill in the field to which the present invention belongs. The words "first", "second" and similar terms used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, words such as "one" or "a" do not indicate a quantity limitation, but rather indicate the existence of at least one. Words such as "connected" or "connected" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship also changes accordingly.

[0021] See Figure 1 , this application provides a DC distribution network fault identification method, comprising:

[0022] S1: Obtain fault power sample data from the distribution automation system in the DC distribution network;

[0023] S2: performing time-frequency decomposition on the fault power sample data using discrete wavelet packet transform to obtain a time-frequency matrix, and converting the time-frequency matrix into a time-frequency spectrum;

[0024] S3: Inputting the time-frequency spectrum into a fault identification network to obtain a fault type of the DC distribution network and a label corresponding to each fault type;

[0025] S4: Use the improved deep residual shrinkage network to extract and identify the key fault features, calculate the probability of mapping the key fault features to the corresponding labels of the fault types of the DC distribution network, and obtain the final fault identification results based on the probability.

[0026] The above-mentioned DC distribution network fault identification method first obtains fault power sample data from the distribution automation system in the DC distribution network. Secondly, the discrete wavelet packet transform is used to perform time-frequency decomposition on the fault power sample data to obtain a time-frequency matrix, and the time-frequency matrix is ​​converted into a pixel matrix of a time-frequency spectrum. Then, labels corresponding to the DC distribution network fault types are constructed. Finally, an improved deep residual shrinkage network is used to extract the key features of the identified faults and implement fault identification and classification. In this way, typical fault problems of the DC distribution network are taken into account, which can help the DC distribution network quickly identify and classify fault types, effectively improving the safety of the power system.

[0027] The following describes in detail the steps of the above-mentioned DC distribution network fault identification method using a complete example:

[0028] 1. Obtain fault power sample data from the distribution automation system in the DC distribution network. The main steps are as follows:

[0029] The historical characteristic quantity dataset of the fault power sample data is classified into a training set and a validation set through the principle of stratified sampling. In one example, the ratio of the training set to the validation set is 4:1. This is for illustrative purposes only and is not intended to be limiting. The fault power sample data is obtained through the distribution automation system in the DC distribution network. The fault power sample data includes but is not limited to the DC transient voltage, current, and power waveforms after historical faults in the DC distribution network.

[0030] 2. Use discrete wavelet packet transform to perform time-frequency decomposition on the fault power sample data to obtain a time-frequency matrix, and then convert the time-frequency matrix into a pixel matrix of the time-frequency spectrum. The main steps are as follows:

[0031] The discrete wavelet packet transform is used to perform time-frequency decomposition on the fault power waveform of the training set to obtain a time-frequency matrix, and then a time-frequency spectrum with an image size of 35×380 pixels is constructed. The time-frequency spectrum is used as the input of the fault identification network.

[0032] 3. Input the time-frequency spectrum into the fault identification network to obtain the fault type of the DC distribution network and the label corresponding to each fault type. The specific steps are as follows:

[0033] The time-frequency spectrum constructed from faulty power sample data is used as the input to the fault identification network. The fault identification network structure is used to establish a mapping relationship between the input and the identification results. The input to the fault identification network is the selection of feature quantities; the output of the fault identification network is the determination of the fault type and the corresponding output labels for different types.

[0034] Therefore, according to the disturbance types that have a significant impact on the DC system, the corresponding labels are set to [L1, L2, L3, L4, L5, L6] to improve the fault identification output results of the deep residual shrinkage network. L1 to L6 represent the probabilities of the first to sixth types of faults. Specifically, L1 is the probability of a positive pole grounding fault; L2 is the probability of a negative pole grounding fault; L3 is the probability of an inter-pole short circuit; L4 is the probability of a heavy load input; L5 is the probability of an AC side symmetrical fault; and L6 is the probability of an asymmetrical fault. The six labels corresponding to the DC distribution network fault types are: the positive pole grounding fault label is [1, 0, 0, 0, 0, 0], the negative pole grounding fault label is [0, 1, 0, 0, 0], the inter-pole short circuit label is [0, 0, 1, 0, 0, 0], the heavy load input label is [0, 0, 0, 1, 0, 0], the AC side symmetrical fault label is [0, 0, 0, 0, 1, 0], and the asymmetrical fault label is [0, 0, 0, 0, 0, 1].

[0035] 4. Use the improved deep residual shrinkage network to extract and identify the key fault features, calculate the probability of mapping the key fault features to the corresponding labels of the DC distribution network fault types, and obtain the final fault identification results based on the probability. The specific steps are as follows:

[0036] (1) The convolutional layer is used to extract the fault characteristics of the DC distribution network. The expression of the convolutional layer operation is:

[0037]

[0038] Where: h j is the output feature of the jth channel of the feature map; X i is the input feature of the i-th channel of the feature map; q ij is the convolution kernel; g j is the bias; M j is the j-th channel set of the feature map.

[0039] (2) The deep residual shrinkage network is improved by introducing an adaptive slope unit into the residual unit to correct the output value of the threshold function, which can improve the network denoising effect. The expression of the new threshold function is:

[0040]

[0041] Where: h t is the output feature; β is the output of the adaptive slope unit; sgn(x) is the sign function, reflecting h j The positive and negative of h j is the input feature, that is, the output feature of the jth channel of the feature map in the convolution layer; θ c is the soft threshold.

[0042] The adaptive slope unit uses the attention mechanism to automatically infer the most appropriate slope. The output β of the adaptive slope unit is:

[0043]

[0044] Where: is the characteristic of the c-th layer neurons.

[0045] (3) The fault feature h is extracted by improving the deep residual shrinkage network t , use Softmax classifier to classify the fault feature h t The probability distribution L = [L1, L2, ..., L6] corresponding to the DC distribution network fault type is mapped, and the category corresponding to the maximum calculated probability is selected as the identification result, thereby realizing the classification of DC distribution network fault identification.

[0046] L=Softmax(Wh t +b)

[0047] Where: W is the weight of the output layer in the improved deep residual shrinkage network; b is the bias.

[0048] (4) Improve the deep residual shrinkage network and convert the focal loss function FL(L n ) introduces a deep residual contraction network to solve the problem of imbalance and misclassification of fault power sample data sets. The focus loss function FL(L n ) is the loss function for fault identification network classification problem to measure the difference between network prediction value and true value, and the focus loss function FL(L n )for:

[0049]

[0050] Where: FL(L n ) is the focal loss function, FL(L n ) is smaller, indicating that the predicted value of the improved deep residual shrinkage network is closer to the actual value; L n In order to improve the prediction probability of the deep residual shrinkage network for the DC distribution network fault type label, L∈[0,1]; L is the actual probability of the DC distribution network fault type label; n is the DC distribution network fault class; γ is the modulation factor index. When γ=0, the focus loss function only solves the tendency problem of fault identification network training; when γ>0, the modulation factor plays a role and focuses on the misclassification situation. The larger γ is, the more attention is focused on the problem of misclassified samples.

[0051] In one example, Table 1 compares the fault identification results of DC distribution networks based on the improved deep residual shrinkage network. The results using different fault identification network types show that the improved deep residual shrinkage network achieves the best identification results, achieving an accuracy of 99.41%, a precision of 97.36%, a recall of 99.32%, and an F1 value of 98.63%. Compared with the deep residual network and the deep residual shrinkage network, the improved deep residual shrinkage network significantly improves fault identification performance.

[0052]

[0053] Where: TP is the number of faulty samples identified correctly; TN is the number of normal samples identified correctly; FP is the number of faulty samples identified incorrectly; FN is the number of normal samples identified incorrectly.

[0054] Table 1 Comparison of DC distribution network fault identification results based on improved deep residual shrinkage network

[0055] Network Type Accuracy / % Accuracy / % Recall / % F1 value / % Deep Residual Network 94.52 93.04 94.86 93.16 Deep Residual Contraction Network 95.34 93.21 94.72 93.86 Improved Deep Residual Contraction Network 99.41 97.36 99.32 98.63

[0056] The present application also provides a DC distribution network fault identification system, comprising:

[0057] A fault information acquisition unit is used to obtain fault power sample data from a distribution automation system in a DC distribution network;

[0058] a fault data processing unit configured to perform time-frequency decomposition of the fault power sample data using discrete wavelet packet transform to obtain a time-frequency matrix, and to convert the time-frequency matrix into a time-frequency spectrum; and further configured to input the time-frequency spectrum into a fault identification network to obtain a fault type of the DC distribution network and a label corresponding to each fault type;

[0059] The fault identification unit is used to extract and identify key fault features using an improved deep residual shrinkage network, calculate the probability of mapping the key fault features to the labels corresponding to the fault types of the DC distribution network, and obtain the final fault identification results based on the probability.

[0060] The above-mentioned DC distribution network fault identification system can implement various embodiments of the above-mentioned DC distribution network fault identification method and achieve the same beneficial effects, which will not be described in detail here.

[0061] This application also provides a DC distribution network fault identification system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-described method are implemented. This DC distribution network fault identification system can implement each embodiment of the above-described DC distribution network fault identification method and achieve the same beneficial effects. These are not further described here.

[0062] The above describes in detail the preferred embodiments of the present invention. It should be understood that those skilled in the art can make numerous modifications and variations based on the concepts of the present invention without inventive effort. Therefore, any technical solutions that can be derived by those skilled in the art through logical analysis, reasoning, or limited experimentation based on the concepts of the present invention and the prior art should be within the scope of protection defined by the claims.

Claims

1. A DC distribution network fault identification method, characterized in that: include: S1: Obtain fault power sample data from the distribution automation system in the DC distribution network; S2: performing time-frequency decomposition on the fault power sample data using discrete wavelet packet transform to obtain a time-frequency matrix, and converting the time-frequency matrix into a time-frequency spectrum; S3: Inputting the time-frequency spectrum into a fault identification network to obtain a fault type of the DC distribution network and a label corresponding to each fault type; S4: Use the improved deep residual shrinkage network to extract and identify the key fault features, calculate the probability of mapping the key fault features to the corresponding labels of the fault types of the DC distribution network, and obtain the final fault identification results based on the probability.

2. The DC distribution network fault identification method according to claim 1, characterized in that: Said S1 comprises: The historical characteristic quantity dataset of fault power sample data is classified into a training set and a validation set through the stratified sampling principle; Fault power sample data is obtained through a distribution automation system in the DC distribution network. The fault power sample data includes DC transient voltage, current and power waveforms after historical faults in the DC distribution network.

3. The DC distribution network fault identification method according to claim 1, characterized in that: The image size of the time-spectrogram in S2 is 35×380 pixels.

4. The DC distribution network fault identification method according to claim 1, characterized in that: The S3 includes: The time-frequency spectrum is used as the input of a fault identification network, and the output of the fault identification network is set with the fault type and the output labels corresponding to the different types; a mapping relationship between the input and the output is established based on the fault identification network; The DC distribution network fault types and their corresponding six labels are as follows: The label for a positive ground fault is [1, 0, 0, 0, 0, 0], the label for a negative ground fault is [0, 1, 0, 0, 0, 0], the label for an inter-pole short circuit is [0, 0, 1, 0, 0, 0], the label for a heavy load input is [0, 0, 0, 1, 0, 0], the label for an AC-side symmetrical fault is [0, 0, 0, 0, 1, 0], and the label for an asymmetrical fault is [0, 0, 0, 0, 0, 1].

5. The DC distribution network fault identification method according to claim 1, characterized in that: The S4 includes: The convolutional layer of the improved deep residual shrinkage network is used to extract the fault characteristics of the DC distribution network. The expression of the convolutional layer operation is as follows: Where: h j is the output feature of the jth channel of the feature map; X i is the input feature of the i-th channel of the feature map; q ij is the convolution kernel; g j is the bias; M j is the j-th channel set of the feature map; The deep residual shrinkage network is improved by introducing an adaptive slope unit into the residual unit to correct the output value of the threshold function. The expression of the threshold function is: Where: h t is the output fault feature; β is the output of the adaptive slope unit; sgn(x) is the sign function, reflecting h j The positive and negative of h j is the input feature, that is, the output feature of the jth channel of the feature map in the convolution layer; θ c is the soft threshold; The adaptive slope unit uses the attention mechanism to automatically infer the optimal slope. The output β of the adaptive slope unit is: Where: is the characteristic of the neurons in layer c; The fault feature h is extracted by improving the deep residual shrinkage network t , use Softmax classifier to classify the fault feature h t Mapping to the label corresponding to the DC distribution network fault type and calculating the probability distribution of the mapping is as follows: L=[L1,L2,…,L6]; Where: L1 is the probability of positive pole grounding fault; L2 is the probability of negative pole grounding fault; L3 is the probability of inter-pole short circuit; L4 is the probability of large load input; L5 is the probability of symmetrical fault on the AC side; L6 is the probability of asymmetrical fault; The fault category corresponding to the maximum calculated probability is selected as the identification result, as follows: L=Softmax(Wh t +b); Where: W is the weight of the output layer in the improved deep residual shrinkage network; b is the bias.

6. The DC distribution network fault identification method according to any one of claims 1 or 5, characterized in that: The improved deep residual shrinkage network includes: using the focus loss function FL(L n ) is the loss function for fault identification network classification problem to measure the difference between network prediction value and true value, and the focus loss function FL(L n )as follows: Where: FL(L n ) is the focal loss function, FL(L n ) is smaller, indicating that the predicted value of the improved deep residual shrinkage network is closer to the actual value; L n In order to improve the prediction probability of the deep residual shrinkage network for the DC distribution network fault type label, L∈[0,1]; L is the actual probability of the DC distribution network fault type label; n is the DC distribution network fault class; γ is the modulation factor index. When γ=0, the focus loss function only solves the tendency problem of fault identification network training; when γ>0, the modulation factor plays a role and focuses on the misclassification situation. The larger γ is, the more attention is focused on the problem of misclassified samples.

7. A DC distribution network fault identification system, characterized in that: include: A fault information acquisition unit is used to obtain fault power sample data from a distribution automation system in a DC distribution network; a fault data processing unit configured to perform time-frequency decomposition of the fault power sample data using discrete wavelet packet transform to obtain a time-frequency matrix, and to convert the time-frequency matrix into a time-frequency spectrum; and further configured to input the time-frequency spectrum into a fault identification network to obtain a fault type of the DC distribution network and a label corresponding to each fault type; The fault identification unit is used to extract and identify key fault features using an improved deep residual shrinkage network, calculate the probability of mapping the key fault features to the labels corresponding to the fault types of the DC distribution network, and obtain the final fault identification results based on the probability.

8. A DC distribution network fault identification system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

Images