DC fault arc detection method and device, and electronic equipment

Abstract

The invention provides a DC fault arc detection method and device and electronic equipment, and belongs to the technical field of discharge detection. The method comprises the steps of obtaining to-be-tested parameters of a direct-current power system; and inputting the to-be-detected parameters into the DC fault arc detection model to obtain a fault detection result output by the DC fault arc detection model. Wherein the model is obtained by training based on a plurality of training sample parameters and a fault detection result label of each training sample parameter, and the plurality of training sample parameters are generated by historical sample parameters and minority class sample parameters of the direct current power system based on a generative adversarial network; the minority class sample parameters are generated by historical sample parameters based on a synthetic minority class oversampling method. According to the method, training sample parameters for training of the DC fault arc detection model are generated by combining the generative adversarial network and the synthetic minority class oversampling method, a data set can be effectively expanded and balanced, and the generalization ability and fault detection accuracy of the DC fault arc detection model are further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of discharge detection, and in particular to a DC fault arc detection method, a DC fault arc detection device, an electronic device, a machine-readable storage medium and a computer program product. Background Art

[0002] In the scenario of variable power consumption, the application scope of DC power distribution has become widespread. As the operation time of the DC power system increases, various problems may occur, such as loose joints, poor contact, aging of wire insulation and device failure, which may cause DC fault arcs at the contact points in the DC power system. Since the DC power supply does not have a zero-crossing characteristic, the protection device cannot effectively detect and identify the DC fault arc. Once a DC fault arc occurs, the arc will continue to burn and release a large amount of energy, seriously affecting the normal operation of the power supply system and may even cause electrical fire accidents.

[0003] The existing solution forms a fault detection model through a machine learning model, then trains the fault detection model, and then determines whether a DC fault arc occurs in the circuit to be detected based on the trained model, and displays the detection results to the operator through a user interface and an alarm system. However, the existing solution has the problem of poor model generalization ability and low fault detection accuracy in the detection of DC fault arcs. Summary of the invention

[0004] The purpose of the embodiments of the present invention is to provide a DC fault arc detection method, a DC fault arc detection device, an electronic device, a machine-readable storage medium and a computer program product to solve the defects of the existing solutions in the detection of DC fault arcs, such as poor model generalization ability and low fault detection accuracy.

[0005] In order to achieve the above object, an embodiment of the present invention provides a DC fault arc detection method, comprising:

[0006] Obtaining parameters to be measured of a DC power system;

[0007] Inputting the measured parameter into a DC fault arc detection model to obtain a fault detection result output by the DC fault arc detection model;

[0008] Among them, the DC fault arc detection model is trained based on multiple training sample parameters and fault detection result labels corresponding to each of the training sample parameters, and the multiple training sample parameters are historical sample parameters and minority class sample parameters of the DC power system generated based on a generative adversarial network; the minority class sample parameters are historical sample parameters generated based on a synthetic minority class oversampling method.

[0009] Optionally, the DC fault arc detection model is trained by the following steps:

[0010] Acquiring a plurality of historical sample parameters of the DC power system;

[0011] Preprocessing all the historical sample parameters to obtain multiple historical feature parameters;

[0012] Based on the synthetic minority class oversampling method, minority class samples are synthesized for all the historical feature parameters to obtain multiple minority class sample parameters;

[0013] Inputting all the historical feature parameters and the minority class sample parameters into the generative adversarial network respectively to obtain the multiple training sample parameters;

[0014] The DC fault arc detection model is trained based on the multiple training sample parameters and the fault detection result label corresponding to each of the training sample parameters.

[0015] Optionally, the synthesizing minority class samples of all the historical feature parameters based on the synthetic minority class oversampling method to obtain multiple minority class sample parameters includes:

[0016] Determine a plurality of minority class parameters among all the historical feature parameters, and determine a nearest neighbor parameter of each of the minority class parameters;

[0017] The minority class sample parameter is generated by performing linear interpolation based on a first minority class parameter and a nearest neighbor parameter of the first minority class parameter; wherein the first minority class parameter is any one of the plurality of minority class parameters.

[0018] Optionally, all the historical feature parameters and the minority class sample parameters are respectively input into the generative adversarial network to obtain the multiple training sample parameters, which are obtained by the following steps:

[0019] Adding noise to each of the historical characteristic parameters to obtain multiple updated historical characteristic parameters;

[0020] Adding noise to each of the minority class sample parameters to obtain multiple updated minority class sample parameters;

[0021] All updated historical feature parameters and all updated minority class sample parameters are respectively input into the generative adversarial network to obtain the multiple training sample parameters.

[0022] Optionally, after inputting all the historical feature parameters and the minority class sample parameters into the generative adversarial network to obtain the plurality of training sample parameters, the method further includes:

[0023] Data enhancement is performed on the multiple training sample parameters by adopting at least one of time shift data output, amplitude scaling data output, and cropping and splicing data output.

[0024] Optionally, the preprocessing of all the historical sample parameters to obtain a plurality of historical feature parameters includes:

[0025] Performing data cleaning on each of the historical sample parameters based on wavelet transform;

[0026] Performing data standardization processing on each of the historical sample parameters after data cleaning;

[0027] Performing data fusion on all the historical sample parameters after data standardization to obtain a fusion signal;

[0028] Performing complete set empirical mode decomposition on the fusion signal to obtain multiple modal components;

[0029] Data feature extraction is performed based on the multiple modal components to obtain the multiple historical feature parameters.

[0030] On the other hand, an embodiment of the present invention further provides a DC fault arc detection device, comprising:

[0031] An acquisition module, used for acquiring parameters to be measured of a DC power system;

[0032] A fault detection module, used for inputting the parameters to be measured into a DC fault arc detection model to obtain a fault detection result output by the DC fault arc detection model;

[0033] Among them, the DC fault arc detection model is trained based on multiple training sample parameters and fault detection result labels corresponding to each of the training sample parameters, and the multiple training sample parameters are historical sample parameters and minority class sample parameters of the DC power system generated based on a generative adversarial network; the minority class sample parameters are historical sample parameters generated based on a synthetic minority class oversampling method.

[0034] Optionally, the DC fault arc detection model is trained by the following steps:

[0035] Acquiring a plurality of historical sample parameters of the DC power system;

[0036] Preprocessing all the historical sample parameters to obtain multiple historical feature parameters;

[0037] Based on the synthetic minority class oversampling method, minority class samples are synthesized for all the historical feature parameters to obtain multiple minority class sample parameters;

[0038] Inputting all the historical feature parameters and the minority class sample parameters into the generative adversarial network respectively to obtain the multiple training sample parameters;

[0039] The DC fault arc detection model is trained based on the multiple training sample parameters and the fault detection result label corresponding to each of the training sample parameters.

[0040] Optionally, the synthesizing minority class samples of all the historical feature parameters based on the synthetic minority class oversampling method to obtain multiple minority class sample parameters includes:

[0041] Determine a plurality of minority class parameters among all the historical feature parameters, and determine a nearest neighbor parameter of each of the minority class parameters;

[0042] The minority class sample parameter is generated by performing linear interpolation based on a first minority class parameter and a nearest neighbor parameter of the first minority class parameter; wherein the first minority class parameter is any one of the plurality of minority class parameters.

[0043] Optionally, all the historical feature parameters and the minority class sample parameters are respectively input into the generative adversarial network to obtain the multiple training sample parameters, which are obtained by the following steps:

[0044] Adding noise to each of the historical characteristic parameters to obtain multiple updated historical characteristic parameters;

[0045] Adding noise to each of the minority class sample parameters to obtain multiple updated minority class sample parameters;

[0046] All updated historical feature parameters and all updated minority class sample parameters are respectively input into the generative adversarial network to obtain the multiple training sample parameters.

[0047] Optionally, after inputting all the historical feature parameters and the minority class sample parameters into the generative adversarial network to obtain the plurality of training sample parameters, the method further includes:

[0048] Data enhancement is performed on the multiple training sample parameters by adopting at least one of time shift data output, amplitude scaling data output, and cropping and splicing data output.

[0049] Optionally, the preprocessing of all the historical sample parameters to obtain a plurality of historical feature parameters includes:

[0050] Performing data cleaning on each of the historical sample parameters based on wavelet transform;

[0051] Performing data standardization processing on each of the historical sample parameters after data cleaning;

[0052] Performing data fusion on all the historical sample parameters after data standardization to obtain a fusion signal;

[0053] Performing complete set empirical mode decomposition on the fusion signal to obtain multiple modal components;

[0054] Data feature extraction is performed based on the multiple modal components to obtain the multiple historical feature parameters.

[0055] On the other hand, the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned DC fault arc detection method when executing the program.

[0056] On the other hand, the present invention further provides a machine-readable storage medium having a computer program stored thereon, and the computer program implements the above-mentioned DC fault arc detection method when executed by a processor.

[0057] On the other hand, the present invention further provides a computer program product, comprising a computer program, wherein the computer program implements the above-mentioned DC fault arc detection method when executed by a processor.

[0058] Through the above technical scheme, the present invention generates training sample parameters for training a DC fault arc detection model by combining a generative adversarial network and a synthetic minority class oversampling method, which can effectively expand and balance the data set, thereby improving the generalization ability and fault detection accuracy of the DC fault arc detection model and improving the effect of DC fault arc detection.

[0059] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0061] Figure 1 It is one of the flow diagrams of the DC fault arc detection method provided by the present invention;

[0062] Figure 2 This is the second flow chart of the DC fault arc detection method provided by the present invention;

[0063] Figure 3 This is the third flow chart of the DC fault arc detection method provided by the present invention;

[0064] Figure 4This is the fourth flow chart of the DC fault arc detection method provided by the present invention;

[0065] Figure 5 This is the fifth flow chart of the DC fault arc detection method provided by the present invention;

[0066] Figure 6 This is the sixth structural schematic diagram of the DC fault arc detection method provided by the present invention;

[0067] Figure 7 It is a structural schematic diagram of a DC fault arc detection device provided by the present invention;

[0068] Figure 8 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0069] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.

[0070] Method Embodiment

[0071] Please refer to Figure 1 The embodiment of the present invention provides a DC fault arc detection method, comprising:

[0072] Step 100: Obtain parameters to be measured of the DC power system.

[0073] The electronic device obtains the parameters to be measured of the DC power system. The parameters to be measured of the DC power system may be various parameters in the DC power system that can reflect the characteristics of the fault arc. For example, the parameters to be measured may be voltage parameters, temperature parameters, and vibration parameters in the DC power system. The voltage parameter refers to the potential difference of the DC power in the DC power system, that is, the potential difference between the positive and negative poles. The temperature in the DC power system refers to the temperature state caused by the heat generated by the various components of the system (such as equipment, cables, busbars, etc.) during operation. The vibration in the DC power system refers to the mechanical vibration generated by the various components of the system (such as motors, generators, etc.) during operation.

[0074] Step 200: input the parameters to be measured into a DC fault arc detection model to obtain a fault detection result output by the DC fault arc detection model.

[0075] The electronic device inputs the parameters to be measured into the DC fault arc detection model to obtain the fault detection result output by the DC fault arc detection model. The DC fault arc detection model is trained based on multiple training sample parameters and the fault detection result label corresponding to each of the training sample parameters. The fault detection result label represents the training sample parameter corresponding to each training sample parameter. The DC fault arc detection model is trained in a supervised manner based on multiple training sample parameters and the fault detection result label corresponding to each of the training sample parameters. The DC fault arc detection model can be implemented using various neural networks. For example, in one embodiment, the DC fault arc detection model is constructed using a BP neural network. The BP neural network is a multi-layer feedforward neural network trained according to the error back propagation algorithm, and multiple training sample parameters are used as learning samples of the neural network, i.e., input feature vectors, to train the DC fault arc detection model.

[0076] In one embodiment, the output value of the DC fault arc detection model constructed by the BP neural network can be three digits, and each digit represents different state information of the circuit. Among them, the first two digits represent the load condition in the circuit: the first two digits are 00 for a single resistance load, 10 for a series resistance-inductance load, and 01 for a parallel resistance-capacitance load. The last digit represents the state of the circuit: when there is a fault arc in the circuit, the last digit is 1; otherwise, it is 2. This model is designed to detect and identify fault arcs in DC power systems. The signal output under normal circuits is 000, without special distinction. By using a BP neural network to construct a DC fault arc detection model, the accuracy, automation and real-time performance of fault detection can be improved, providing reliable support for circuit operation and maintenance.

[0077] The multiple training sample parameters are historical sample parameters and minority class sample parameters of the DC power system generated based on Generative Adversarial Networks (GANs). The minority class sample parameters are generated by the historical sample parameters based on the synthetic minority class oversampling method (SMOTE, Synthetic Minority Over-Sampling Technique). The historical sample parameters may include various parameters of the DC power system in the past period of time (5 days). For example, the historical sample parameters may include historical voltage parameters, historical temperature parameters, and historical vibration parameters. In order to improve the coverage of the data, in one embodiment, the historical sample parameters may include historical voltage parameters, historical temperature parameters, and historical vibration parameters.

[0078] The embodiment of the present invention first generates minority class sample parameters from historical sample parameters using a synthetic minority class oversampling method. Then, the historical sample parameters and minority class sample parameters are respectively input into a generative adversarial network to generate new data samples for model training. The use of the synthetic minority class oversampling method and the generative adversarial network for data enhancement effectively expands the diversity and balance of the data set, and improves the generalization ability of the model and the ability to recognize minority class samples.

[0079] The present invention generates training sample parameters for training a DC fault arc detection model by combining a generative adversarial network and a synthetic minority class oversampling method, which can effectively expand and balance the data set, thereby improving the generalization ability and fault detection accuracy of the DC fault arc detection model and improving the effect of DC fault arc detection.

[0080] For other aspects of the embodiments of the present invention, please refer to Figure 2 The DC fault arc detection model is trained by the following steps:

[0081] Step 10: Acquire a plurality of historical sample parameters of the DC power system.

[0082] In one embodiment, the electronic device may obtain historical voltage parameters, historical temperature parameters, and historical vibration parameters of the DC power system in the past period of time (5 days) as a plurality of historical sample parameters.

[0083] Step 20: pre-process all the historical sample parameters to obtain a plurality of historical feature parameters.

[0084] In DC fault arc detection, the collected historical sample parameters need to be preprocessed and enhanced to improve data quality and detection performance. For example, electronic equipment can perform data cleaning on historical voltage parameters, historical temperature parameters, and historical vibration parameters to retain useful fault information. Then, data standardization is used to normalize the data to a uniform scale, eliminating the dimensional differences between different features, thereby extracting data features and finally obtaining multiple historical feature parameters.

[0085] Step 30: synthesize minority class samples for all the historical feature parameters based on a synthetic minority class oversampling method to obtain a plurality of minority class sample parameters.

[0086] The electronic device generates minority class samples for all the historical feature parameters based on the synthetic minority class oversampling method to obtain multiple minority class sample parameters. Specifically, the principle of the synthetic minority class oversampling method is that for each minority class sample x i , find its nearest neighbor sample x NN , generate new samples y by linear interpolation i .

[0087] Step 30: synthesize minority class samples for all the historical feature parameters based on the synthetic minority class oversampling method to obtain multiple minority class sample parameters, specifically including:

[0088] Step 31: determining a plurality of minority class parameters among all the historical feature parameters, and determining a nearest neighbor parameter of each of the minority class parameters;

[0089] Step 32: Perform linear interpolation based on the first minority class parameter and the nearest neighbor parameter of the first minority class parameter to generate the minority class sample parameter.

[0090] The electronic device determines a plurality of minority class parameters x from among all the historical feature parameters i The minority class parameters can be the minority class parameters with low occurrence frequency, or the minority class parameters that deviate far from the historical average level. For example, when the 10 data are 0, 0.4, 0.5, 0.5, 0.6, 0.7, 0.4, 0.5, 0.6, 1. The minority class samples can be 0, 0.7 or 1 with low occurrence frequency, or 0 or 1 that deviate far from the historical average level.

[0091] In the case of determining multiple minority class parameters, the electronic device further determines the nearest neighbor parameter x of each of the minority class parameters. NN Then, linear interpolation is performed based on the first minority class parameter and the nearest neighbor parameter of the first minority class parameter to generate the minority class sample parameter. The first minority class parameter is any minority class parameter among the multiple minority class parameters. That is, each minority class parameter executes the above method to obtain the minority class sample parameter, thereby forming multiple minority class sample parameters.

[0092] Based on the first minority class parameter and the nearest neighbor parameter of the first minority class parameter, linear interpolation is performed to generate the minority class sample parameter, which can be expressed by the following formula:

[0093] y i =x i +λ(x NN -x i );

[0094] Among them, x i is the first minority class parameter, x NN is the nearest neighbor parameter of the first minority class parameter, λ is a random number, 0≤λ≤1, y i is the minority class sample parameter.

[0095] Therefore, the embodiment of the present invention generates new minority class samples by synthesizing the minority class oversampling method to achieve a balanced data set.

[0096] Step 40: input all the historical feature parameters and the minority class sample parameters into the generative adversarial network respectively to obtain the multiple training sample parameters.

[0097] The electronic device generates new data samples based on all the historical feature parameters and the minority class sample parameters through a generative adversarial network (GANs). During the training of the generative adversarial network, the generator G of the generative adversarial network accepts random noise z as input to generate simulated data G(z). The discriminator D of the generative adversarial network is trained by judging whether the input simulated data G(z) is real data or generated data, and its goal is to optimize the loss function:

[0098]

[0099] In the above formula, G is a generator that generates simulated data G(z) by accepting random noise z as input; D is a discriminator that determines whether the input data is real data or generated data, and outputs a probability value indicating that the input data is real data; V(D,G) is the objective function in the generative adversarial network (GANs); p data (y) is the distribution of real data, which represents the probability distribution of real data sample y; p z (z) is the noise distribution, which represents the probability distribution of random noise z; D(y) represents the probability value that the discriminator judges the real data y as real; Indicates that in the real data distribution p data (y) is the desired operation under Indicates that in the noise distribution p z (z) The expected operation under

[0100] When the loss function is satisfied, the output generates a new data sample z i , that is, the training sample parameters are obtained.

[0101] After the generative adversarial network is trained, Gaussian noise is added to the historical feature parameters and the minority class sample parameters respectively, and then the historical feature parameters and the minority class sample parameters after adding Gaussian noise are input into the generative adversarial network to obtain the multiple training sample parameters.

[0102] Specifically, in step 40, all the historical feature parameters and the minority class sample parameters are respectively input into the generative adversarial network to obtain the multiple training sample parameters, which are obtained by the following steps:

[0103] Step 41, adding noise to each of the historical characteristic parameters to obtain multiple updated historical characteristic parameters;

[0104] Step 42: adding noise to each of the minority class sample parameters to obtain multiple updated minority class sample parameters;

[0105] Step 43: input all updated historical feature parameters and all updated minority class sample parameters into the generative adversarial network respectively to obtain the multiple training sample parameters.

[0106] The electronic device adds noise to each of the historical feature parameters to obtain multiple updated historical feature parameters, and adds noise to each of the minority class sample parameters to obtain multiple updated minority class sample parameters. The process of adding Gaussian noise is represented by the following formula:

[0107] z′ i =y i +∈;

[0108] Among them, Z i 'Indicates updating historical feature parameters or updating minority class sample parameters. i Represents historical feature parameters or minority class sample parameters. ∈ represents Gaussian noise that follows a normal distribution.

[0109] The plurality of training sample parameters are obtained by inputting all updated historical feature parameters and all updated minority class sample parameters into the generative adversarial network. The embodiment of the present invention uses the generative adversarial network to generate new data samples, thereby expanding the diversity and coverage of the data set.

[0110] Therefore, the embodiment of the present invention utilizes the synthetic minority class oversampling method and the generative adversarial network for data enhancement, which effectively expands the diversity and balance of the data set and improves the generalization ability of the model and the recognition ability of minority class samples.

[0111] Step 60: Training the DC arc fault detection model based on the multiple training sample parameters and the fault detection result label corresponding to each of the training sample parameters.

[0112] Please refer to Figure 3 The electronic device inputs the training sample parameters as training feature vectors into the DC fault arc detection model to obtain the fault detection result of the BP neural network. The loss function is calculated based on the fault detection result and the fault detection result label corresponding to the training sample parameters. When the loss function meets the training error condition, the training of the BP neural network is completed, so that the feature vector formed by the measured parameters of the DC power system can be predicted. The loss function can be calculated using various known loss functions, such as the cross entropy loss function.

[0113] For other aspects of the embodiments of the present invention, please refer to Figure 4, step 40, inputting all the historical feature parameters and the minority class sample parameters into the generative adversarial network respectively, and obtaining the plurality of training sample parameters, further comprising:

[0114] Step 50: Perform data enhancement on the multiple training sample parameters by adopting at least one of time shift data output, amplitude scaling data output, and cropping and splicing data output.

[0115] Among them, the time offset data output refers to the time offset of the time series data to generate a new training sample parameter z shifted It is expressed by the following formula:

[0116] z shifted (t) = z′(t+Δt);

[0117] Where Δt is a random time offset. Z'(t) represents the function form of the original time series data.

[0118] The amplitude scaling data output refers to the random amplitude scaling of the training sample parameters to generate a new scaled data set, including the training sample parameters z after amplitude scaling. scaled :

[0119] z scaled =αz′ i ;

[0120] Here, α is a random scaling factor.

[0121] The cropping and splicing data output refers to the process of cropping long time series data into short segments and re-splicing them to form a cropped and spliced ​​data set. The data set contains the reorganized time series samples to generate new data. cropped :

[0122] z cropped =[z′(t 1 , t 2 ), z′(t 3 , t 4 )]

[0123] Among them, t 1 ,t 2 ,t 3 ,t 4 is the clipping point of the time series.

[0124] It should be noted that the embodiment of the present invention can perform data enhancement on the multiple training sample parameters by one or more of the following methods: time offset data output, amplitude scaling data output, and cropping and splicing data output. In order to improve the input data quality of the DC fault arc detection model, in one embodiment, the embodiment of the present invention can perform data enhancement on the multiple training sample parameters by three methods: time offset data output, amplitude scaling data output, and cropping and splicing data output. Through the above data enhancement method, the input data quality of the DC fault arc detection model can be significantly improved, and the diversity and coverage of the data can be enhanced, thereby improving the training effect of the model and the accuracy of fault detection.

[0125] For other aspects of the embodiments of the present invention, please refer to Figure 5 , step 20, preprocessing all the historical sample parameters to obtain multiple historical feature parameters, including:

[0126] Step 21: Perform data cleaning on each of the historical sample parameters based on wavelet transform.

[0127] In order to improve data quality and detection performance, electronic devices can use wavelet transform noise reduction technology to remove noise components in historical sample parameters and retain useful fault information.

[0128] Specifically, data cleaning is performed on each of the historical sample parameters based on wavelet transform.

[0129] The following formula represents:

[0130] W j,k =∫x(t)ψ j,k (t)dt;

[0131] Among them, W j,k is the wavelet basis function, j and k represent the scale and position parameters respectively, x(t) represents the original signal or historical sample parameter at time t, and is the signal that needs to be transformed by wavelet. j,k ( represents the wavelet basis function, which is the transformation function at different scales j and positions k. By changing the values ​​of j and k, the wavelet basis at different time positions and frequency scales can be obtained to analyze the local characteristics of the signal. Then the wavelet coefficients are thresholded to remove the noise coefficients and obtain the denoised wavelet coefficients

[0132]

[0133] The threshold parameter λ is usually taken as Here σ is the standard deviation of the noise and n is the length of the signal. Finally, the denoised wavelet coefficients are reconstructed into the signal by inverse wavelet transform.

[0134] in, is the denoised signal. Through the above wavelet denoising process, the noise in multiple historical sample parameters can be effectively removed, the fault characteristics can be retained, and the data quality can be improved.

[0135] Step 22: Perform data standardization on each of the historical sample parameters after data cleaning.

[0136] In DC fault arc detection, data standardization is a key step to ensure that the input data is on the same scale, which helps to improve the convergence speed and stability of the model. Data standardization converts each of the historical sample parameters into a standard normal distribution with a mean of 0 and a standard deviation of 1. Assume that the historical sample parameter is x i (where i = 1, 2, ..., n; n represents the number of samples of historical sample parameters), first calculate the mean μ and standard deviation σ of multiple historical sample parameters:

[0137]

[0138] Then, each historical sample parameter is standardized to obtain the standardized historical sample parameter z i :

[0139]

[0140] Among them, μ represents the mean of multiple historical sample parameters, σ represents the standard deviation of multiple historical sample parameters, and x i represents the original historical sample parameters, z i Represents the standardized historical sample parameters. Through this standardization process, each historical sample parameter z i They are all converted into standard normal distribution data with a mean of 0 and a standard deviation of 1, thereby eliminating the dimensional differences between different features and improving the convergence speed and stability of the model.

[0141] Step 23: perform data fusion on all the historical sample parameters after data standardization to obtain a fused signal.

[0142] In DC fault arc detection, multi-source data fusion is an important step to improve detection accuracy. Multi-source data fusion combines data from different sensors to obtain comprehensive features that can capture fault characteristics more comprehensively. Assume that the data obtained from different sensors and after data standardization is x i (t), y i (t), z i(t), respectively represent the voltage standardization signal, temperature standardization signal and vibration standardization signal, where i represents the i-th time point. The data is fused by weighted average method to obtain the fused signal f i (t):

[0143] f i (t) = αx i (t)+βy i (t)+γz i (t);

[0144] Among them, α, β and γ are weighted coefficients, which represent the weight of each sensor data in the fusion process, and must satisfy α + β + γ = 1. Through this multi-source data fusion method, the information of different sensors can be integrated to improve the accuracy and robustness of fault detection, thereby more comprehensively reflecting the characteristics of DC fault arcs.

[0145] Step 24: Perform complete set empirical mode decomposition on the fusion signal to obtain multiple modal components.

[0146] Although the existing fault arc detection is relatively mature in AC fault arc detection, it has certain defects in DC fault arc detection. DC fault arc has randomness and uncertainty, which leads to complex detection process, long data processing time and low detection accuracy. The decomposition of DC fault arc signal is a key step in determining the detection speed and detection accuracy.

[0147] For example, after decomposing the signal, the traditional wavelet decomposition obtains the specific frequency band signal in the original signal, but the decomposed signal has little correlation with the original signal, which affects the accuracy of subsequent feature extraction and reduces the recognition accuracy of the model. While the empirical mode decomposition is simpler than the traditional wavelet decomposition in terms of algorithm operation and program writing, it is easy to have mode aliasing in the intrinsic mode function components after decomposition, which seriously affects the detection efficiency of DC fault arc.

[0148] The embodiment of the present invention decomposes the fused signal through the complete set empirical mode decomposition method to obtain multiple modal components, which can reduce noise interference, solve the problem of modal aliasing, improve decomposition efficiency, and make it easier to obtain the inherent frequency band signal in the original signal. Figure 6 The steps of completing the empirical mode decomposition of the set in the embodiment of the present invention are as follows:

[0149] 1) Obtain the intrinsic mode function IMF 1 :

[0150] Get the first IMF 1 The method is the same as the empirical mode decomposition, and its calculation formula is as follows:

[0151]

[0152] Among them, IMF i1 represents the first intrinsic mode function obtained by decomposing the i-th experimental data, i is the number of experiments (i=1,2,…,M), IMF 1 (n) represents the first natural mode component corresponding to the fault signal f(n).

[0153] 2) Calculate the residual signal h 1 (n):

[0154] Assuming that the collected fault signal is f(n), the calculation formula of the remaining signal after the first step is as follows:

[0155] h 1 (n) = f(n) - IMF 1 (n);

[0156] Among them, h 1 (n) represents the residual signal after the first decomposition; f(n) represents the extracted fault signal; IMF 1 (n) represents the first natural mode component corresponding to the fault signal f(n).

[0157] 3) Obtain the intrinsic mode function IMF 2 :

[0158] E K is the jth intrinsic mode function generated by the empirical mode decomposition method, σ i The Gaussian white noise added in the calculation of complete set empirical mode decomposition satisfies the normal distribution. When the decomposition calculation obtains the first component of empirical mode decomposition, the calculation stops and the second intrinsic mode function is calculated. The calculation formula is as follows:

[0159] Among them, IMF 2 (n) represents the second natural mode component corresponding to the fault signal f(n). i = 1, 2, ..., m.

[0160] Where f(n) is updated in the third step as:

[0161] f(n)=E 1 (h 1 (n)+E 1 σ i (n)); where E 1 is the first intrinsic mode function generated by the complete ensemble empirical mode decomposition method; σ i (n) is the Gaussian white noise added in the i-th experiment of complete ensemble empirical mode decomposition, which satisfies N(0,1) distribution, i=1, 2,…,M.

[0162] 4) Residual signal judgment:

[0163] The remaining signal is constantly judged in the program. When the stop condition is not met, the nth remaining signal and the n+1th intrinsic mode function are calculated. The calculation formula is as follows:

[0164] h n (n) = h n-1 (n)-IMF n (n)

[0165]

[0166] Among them, h n (n) represents the residual signal after the nth decomposition; h n-1 (n) represents the residual signal after the n-1th decomposition; f(n) represents the extracted fault signal; IMF n (n) represents the nth intrinsic modal component of the corresponding fault signal f(n); IMF n+1 (n) represents the n+1th natural mode component corresponding to the fault signal f(n).

[0167] When the stop condition is met, the final remaining signal and the signal to be processed are output, and the calculation formula is as follows:

[0168]

[0169] The stopping condition is h n The number of extreme points of (n) is less than 2. M is the number of IMFs at the end of the decomposition calculation. i represents the i-th intrinsic mode component, IMF n represents the nth natural modal component, n=1, 2, ..., m; f(n) represents the fault signal; H(n) represents the final residual signal corresponding to the fault signal.

[0170] The embodiment of the present invention improves the extraction and decomposition efficiency of fault signals and significantly improves the detection accuracy by combining multi-source data fusion and empirical mode decomposition (EMD) technology. In addition, in view of the existing problems of DC fault arc detection and the characteristics of DC fault arc, the mode decomposition technology is used to reduce the complexity of data processing, improve the efficiency and accuracy of signal decomposition, and at the same time, combined with subsequent data enhancement technology, by generating more diverse and high-quality training data, the real-time and accuracy of DC fault arc detection are further improved. Data enhancement technology includes multi-source data fusion, data cleaning and noise reduction, data perturbation and generative adversarial networks. The application of these technologies can significantly improve the model's recognition ability and generalization performance for fault signals, thereby solving the application difficulties of existing detection methods in complex variable power consumption scenarios.

[0171] Step 25: extract data features based on the multiple modal components to obtain the multiple historical feature parameters.

[0172] In DC fault arc detection, feature extraction is an important step in identifying and distinguishing faults. The embodiment of the present invention can more accurately detect faults by extracting key features of the signal. Feature extraction includes time domain features, frequency domain features and time-frequency features. First, time domain features such as mean μ, standard deviation σ, skewness γ 1 and kurtosis γ 2 is the original signal x i Directly calculated in:

[0173]

[0174] Where n represents the number of samples, x i represents the i-th modal component.

[0175] Frequency domain features are extracted through Fourier transform, such as spectrum energy E:

[0176]

[0177] Among them, X(f k ) indicates that the signal is at frequency f k The Fourier transform coefficient at , N represents the number of frequency components.

[0178] The time-frequency features can be calculated by short-time Fourier transform (STFT) to obtain a joint representation of time and frequency:

[0179]

[0180] Among them, X(t,f) is the time-frequency feature, x(τ) is the original signal, w(τ-t) is the window function, f is the frequency, and t is the time. By extracting these features, the important information in the signal can be effectively captured and the accuracy of DC fault arc detection can be improved.

[0181] Therefore, the embodiment of the present invention pre-processes and decomposes the historical sample parameters collected in the DC power system. First, the noise interference is removed by wavelet noise reduction technology, and then the data is standardized to ensure signal consistency. Next, through multi-source data fusion technology, historical voltage parameters, severe temperature parameters, historical vibration parameters and other data are combined to enhance the comprehensiveness of the signal. Subsequently, the enhanced signal is decomposed into several intrinsic mode functions (IMFs) using the complete set empirical mode decomposition technology. New minority class samples are generated by the synthetic minority class oversampling method to balance the data set, and then new data samples are generated using the generative adversarial network to expand the diversity and coverage of the data set. Finally, the feature vector is transmitted to the BP neural network for fault arc identification.

[0182] The embodiment of the present invention combines data enhancement and complete set empirical mode decomposition to form a DC fault arc detection method, the core of which is the combination of data enhancement and signal decomposition technology. Through wavelet denoising and data standardization preprocessing, the signal is enhanced by using multi-source data fusion technology, and then the signal is decomposed into several intrinsic mode functions (IMFs) through complete set empirical mode decomposition to solve the problem of modal aliasing. Diverse data samples are generated by synthetic minority class oversampling methods and generative adversarial network technology to expand the coverage of the data set. The enhanced data features are used for fault identification through BP neural network, thereby improving the detection capability and accuracy and shortening the detection time.

[0183] The embodiment of the present invention proposes a DC fault arc detection method combining data enhancement and signal decomposition for DC fault arcs in DC power systems, which can effectively improve the accuracy and real-time performance of fault detection. By combining multi-source data fusion and complete set empirical mode decomposition technology, the extraction and decomposition efficiency of fault signals are improved, and the detection accuracy is significantly improved. Secondly, data enhancement is performed using generative adversarial networks and synthetic minority class oversampling methods, which effectively expands the diversity and balance of data sets, improves the generalization ability of the model and the recognition ability of minority class samples. In addition, the added data preprocessing steps, including wavelet denoising and data standardization processing, ensure the high quality and consistency of the input data. Through these improvements, the embodiment of the present invention not only overcomes the defects of the prior art in DC fault arc detection, but also achieves the high efficiency and accuracy of fault arc identification, providing important guarantees for the safe and stable operation of DC power systems.

[0184] Device Embodiment

[0185] Please refer to Figure 7 On the other hand, an embodiment of the present invention further provides a DC fault arc detection device, comprising:

[0186] An acquisition module 701 is used to acquire parameters to be measured of a DC power system;

[0187] A fault detection module 702 is used to input the parameters to be measured into a DC fault arc detection model to obtain a fault detection result output by the DC fault arc detection model;

[0188] Among them, the DC fault arc detection model is trained based on multiple training sample parameters and fault detection result labels corresponding to each of the training sample parameters, and the multiple training sample parameters are historical sample parameters and minority class sample parameters of the DC power system generated based on a generative adversarial network; the minority class sample parameters are historical sample parameters generated based on a synthetic minority class oversampling method.

[0189] Optionally, the DC fault arc detection model is trained by the following steps:

[0190] Acquiring a plurality of historical sample parameters of the DC power system;

[0191] Preprocessing all the historical sample parameters to obtain multiple historical feature parameters;

[0192] Based on the synthetic minority class oversampling method, minority class samples are synthesized for all the historical feature parameters to obtain multiple minority class sample parameters;

[0193] Inputting all the historical feature parameters and the minority class sample parameters into the generative adversarial network respectively to obtain the multiple training sample parameters;

[0194] The DC fault arc detection model is trained based on the multiple training sample parameters and the fault detection result label corresponding to each of the training sample parameters.

[0195] Optionally, the synthesizing minority class samples of all the historical feature parameters based on the synthetic minority class oversampling method to obtain multiple minority class sample parameters includes:

[0196] Determine a plurality of minority class parameters among all the historical feature parameters, and determine a nearest neighbor parameter of each of the minority class parameters;

[0197] The minority class sample parameter is generated by performing linear interpolation based on a first minority class parameter and a nearest neighbor parameter of the first minority class parameter; wherein the first minority class parameter is any one of the plurality of minority class parameters.

[0198] Optionally, all the historical feature parameters and the minority class sample parameters are respectively input into the generative adversarial network to obtain the multiple training sample parameters, which are obtained by the following steps:

[0199] Adding noise to each of the historical characteristic parameters to obtain multiple updated historical characteristic parameters;

[0200] Adding noise to each of the minority class sample parameters to obtain multiple updated minority class sample parameters;

[0201] All updated historical feature parameters and all updated minority class sample parameters are respectively input into the generative adversarial network to obtain the multiple training sample parameters.

[0202] Optionally, the device further comprises:

[0203] The data enhancement module is used to perform data enhancement on the multiple training sample parameters by adopting at least one of time offset data output, amplitude scaling data output, and cropping and splicing data output.

[0204] Optionally, the preprocessing of all the historical sample parameters to obtain a plurality of historical feature parameters includes:

[0205] Performing data cleaning on each of the historical sample parameters based on wavelet transform;

[0206] Performing data standardization processing on each of the historical sample parameters after data cleaning;

[0207] Performing data fusion on all the historical sample parameters after data standardization to obtain a fusion signal;

[0208] Performing complete set empirical mode decomposition on the fusion signal to obtain multiple modal components;

[0209] Data feature extraction is performed based on the multiple modal components to obtain the multiple historical feature parameters.

[0210] The DC fault arc detection device includes a processor and a memory. The acquisition module 701, the fault detection module 702, etc. are stored in the memory as program units. The processor executes the program units stored in the memory to implement corresponding functions.

[0211] The processor includes a kernel, which calls the corresponding program unit from the memory. There can be one or more kernels.

[0212] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0213] Figure 8 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 8As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication interface 820 and the memory 830 communicate with each other through the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute the DC fault arc detection method, which includes: obtaining the parameters to be measured of the DC power system; inputting the parameters to be measured into the DC fault arc detection model to obtain the fault detection result output by the DC fault arc detection model; wherein the DC fault arc detection model is trained based on multiple training sample parameters and the fault detection result label corresponding to each of the training sample parameters, wherein the multiple training sample parameters are the historical sample parameters of the DC power system and the minority class sample parameters generated based on the generative adversarial network; the minority class sample parameters are the historical sample parameters generated based on the synthetic minority class oversampling method.

[0214] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0215] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a machine-readable storage medium. When the computer program is executed by a processor, the computer can execute a DC fault arc detection method, which includes: obtaining parameters to be measured of a DC power system; inputting the parameters to be measured into a DC fault arc detection model to obtain a fault detection result output by the DC fault arc detection model; wherein the DC fault arc detection model is trained based on multiple training sample parameters and fault detection result labels corresponding to each of the training sample parameters, and the multiple training sample parameters are historical sample parameters of the DC power system and minority class sample parameters generated based on a generative adversarial network; the minority class sample parameters are historical sample parameters generated based on a synthetic minority class oversampling method.

[0216] On the other hand, the present invention also provides a machine-readable storage medium having a computer program stored thereon, which is implemented when a processor executes a DC fault arc detection method, the method comprising: obtaining parameters to be measured of a DC power system; inputting the parameters to be measured into a DC fault arc detection model to obtain a fault detection result output by the DC fault arc detection model; wherein the DC fault arc detection model is trained based on multiple training sample parameters and fault detection result labels corresponding to each of the training sample parameters, the multiple training sample parameters are historical sample parameters of the DC power system and minority class sample parameters generated based on a generative adversarial network; the minority class sample parameters are generated by the historical sample parameters based on a synthetic minority class oversampling method.

[0217] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0218] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0219] 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 aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A DC fault arc detection method, characterized in that: include: Obtaining parameters to be measured of the DC power system; Inputting the measured parameter into a DC fault arc detection model to obtain a fault detection result output by the DC fault arc detection model; Among them, the DC fault arc detection model is trained based on multiple training sample parameters and fault detection result labels corresponding to each of the training sample parameters, and the multiple training sample parameters are historical sample parameters and minority class sample parameters of the DC power system generated based on a generative adversarial network; the minority class sample parameters are historical sample parameters generated based on a synthetic minority class oversampling method.

2. The DC fault arc detection method according to claim 1, characterized in that: The DC fault arc detection model is trained by the following steps: Acquiring a plurality of historical sample parameters of the DC power system; Preprocessing all the historical sample parameters to obtain multiple historical feature parameters; Based on the synthetic minority class oversampling method, minority class samples are synthesized for all the historical feature parameters to obtain multiple minority class sample parameters; Inputting all the historical feature parameters and the minority class sample parameters into the generative adversarial network respectively to obtain the multiple training sample parameters; The DC fault arc detection model is trained based on the multiple training sample parameters and the fault detection result label corresponding to each of the training sample parameters.

3. The DC fault arc detection method according to claim 2, characterized in that: The method based on synthetic minority class oversampling performs minority class sample synthesis on all the historical feature parameters to obtain multiple minority class sample parameters, including: Determine a plurality of minority class parameters among all the historical feature parameters, and determine a nearest neighbor parameter of each of the minority class parameters; The minority class sample parameter is generated by performing linear interpolation based on a first minority class parameter and a nearest neighbor parameter of the first minority class parameter; wherein the first minority class parameter is any one of the plurality of minority class parameters.

4. The DC fault arc detection method according to claim 2, characterized in that: The steps of inputting all the historical feature parameters and the minority class sample parameters into the generative adversarial network to obtain the multiple training sample parameters are as follows: Adding noise to each of the historical characteristic parameters to obtain multiple updated historical characteristic parameters; Adding noise to each of the minority class sample parameters to obtain multiple updated minority class sample parameters; All updated historical feature parameters and all updated minority class sample parameters are respectively input into the generative adversarial network to obtain the multiple training sample parameters.

5. The DC fault arc detection method according to claim 2, characterized in that: After inputting all the historical feature parameters and the minority class sample parameters into the generative adversarial network to obtain the plurality of training sample parameters, the method further includes: Data enhancement is performed on the multiple training sample parameters by adopting at least one of time shift data output, amplitude scaling data output, and cropping and splicing data output.

6. The DC fault arc detection method according to claim 2, characterized in that: The preprocessing of all the historical sample parameters to obtain a plurality of historical characteristic parameters includes: Performing data cleaning on each of the historical sample parameters based on wavelet transform; Performing data standardization processing on each of the historical sample parameters after data cleaning; Performing data fusion on all the historical sample parameters after data standardization to obtain a fusion signal; Performing complete set empirical mode decomposition on the fusion signal to obtain multiple modal components; Data feature extraction is performed based on the multiple modal components to obtain the multiple historical feature parameters.

7. A DC fault arc detection device, characterized in that: include: An acquisition module, used for acquiring parameters to be measured of a DC power system; A fault detection module, used for inputting the parameters to be measured into a DC fault arc detection model to obtain a fault detection result output by the DC fault arc detection model; Among them, the DC fault arc detection model is trained based on multiple training sample parameters and fault detection result labels corresponding to each of the training sample parameters, and the multiple training sample parameters are historical sample parameters and minority class sample parameters of the DC power system generated based on a generative adversarial network; the minority class sample parameters are historical sample parameters generated based on a synthetic minority class oversampling method.

8. The DC fault arc detection device according to claim 7, characterized in that: The DC fault arc detection model is trained by the following steps: Acquiring a plurality of historical sample parameters of the DC power system; Preprocessing all the historical sample parameters to obtain multiple historical feature parameters; Based on the synthetic minority class oversampling method, minority class samples are synthesized for all the historical feature parameters to obtain multiple minority class sample parameters; Inputting all the historical feature parameters and the minority class sample parameters into the generative adversarial network respectively to obtain the multiple training sample parameters; The DC fault arc detection model is trained based on the multiple training sample parameters and the fault detection result label corresponding to each of the training sample parameters.

9. The DC fault arc detection device according to claim 8, characterized in that: The method based on synthetic minority class oversampling performs minority class sample synthesis on all the historical feature parameters to obtain multiple minority class sample parameters, including: Determine a plurality of minority class parameters among all the historical feature parameters, and determine a nearest neighbor parameter of each of the minority class parameters; The minority class sample parameter is generated by performing linear interpolation based on a first minority class parameter and a nearest neighbor parameter of the first minority class parameter; wherein the first minority class parameter is any one of the plurality of minority class parameters.

10. The DC fault arc detection device according to claim 8, characterized in that: The steps of inputting all the historical feature parameters and the minority class sample parameters into the generative adversarial network to obtain the multiple training sample parameters are as follows: Adding noise to each of the historical characteristic parameters to obtain multiple updated historical characteristic parameters; Adding noise to each of the minority class sample parameters to obtain multiple updated minority class sample parameters; All updated historical feature parameters and all updated minority class sample parameters are respectively input into the generative adversarial network to obtain the multiple training sample parameters.

11. The DC fault arc detection device according to claim 8, characterized in that: After inputting all the historical feature parameters and the minority class sample parameters into the generative adversarial network to obtain the plurality of training sample parameters, the method further includes: Data enhancement is performed on the multiple training sample parameters by adopting at least one of time shift data output, amplitude scaling data output, and cropping and splicing data output.

12. The DC fault arc detection device according to claim 8, characterized in that: The preprocessing of all the historical sample parameters to obtain a plurality of historical characteristic parameters includes: Performing data cleaning on each of the historical sample parameters based on wavelet transform; Performing data standardization processing on each of the historical sample parameters after data cleaning; Performing data fusion on all the historical sample parameters after data standardization to obtain a fusion signal; Performing complete set empirical mode decomposition on the fusion signal to obtain multiple modal components; Data feature extraction is performed based on the multiple modal components to obtain the multiple historical feature parameters.

13. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the DC fault arc detection method according to any one of claims 1 to 6 is implemented.

14. A machine-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the DC fault arc detection method according to any one of claims 1 to 6 is implemented.

15. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the DC fault arc detection method according to any one of claims 1 to 6 is implemented.

Images