Feature selection method, device, computer equipment and storage medium

Abstract

The present application relates to a feature selection method, apparatus, computer device, and storage medium. Initial features of a target arc signal are obtained, sorted by importance score, and at least one candidate feature combination is determined from the initial features. Each candidate feature combination is input into a preset accuracy prediction model to obtain a first fault identification accuracy for each candidate feature combination. Finally, a target feature combination is determined based on the first fault identification accuracy for each candidate feature combination. This method can quickly and accurately screen out suitable features, and the selected suitable features can achieve the highest accuracy arc fault identification result using the fewest features.

Description

Technical Field

[0001] The present application relates to the technical field of data feature processing, and in particular to a feature selection method, apparatus, computer equipment, and storage medium. Background Art

[0002] Feature selection plays an increasingly important role in many fields such as data mining, machine learning, and pattern recognition. Feature selection can reduce the complexity of research problems in various fields and improve the prediction accuracy, robustness, and interpretability of learning algorithms.

[0003] Feature selection involves selecting a subset of features from the original feature set that best reflects a certain evaluation criterion. Choosing too few features during feature selection results in poor modeling results and limited interpretability, while selecting too many features results in slower execution.

[0004] Therefore, how to quickly and accurately screen out suitable features has become a technical problem that needs to be solved urgently. Summary of the Invention

[0005] Based on this, it is necessary to provide a feature selection method, device, computer equipment and storage medium to address the above technical problems, which can quickly and accurately screen out suitable features. The selected suitable features can achieve the highest accuracy arc fault identification results with the least features.

[0006] In a first aspect, an embodiment of the present application provides a feature selection method, the method comprising:

[0007] Acquire initial features of a target arc signal; the initial features include multiple types of features;

[0008] sorting the features of the multiple types according to their importance scores, and determining at least one candidate feature combination from the multiple types of features;

[0009] Inputting each candidate feature combination into a preset accuracy prediction model to obtain the first fault identification accuracy of each candidate feature combination;

[0010] The target feature combination is determined according to the first fault identification accuracy of each candidate feature combination.

[0011] In one embodiment, obtaining initial characteristics of a target arc signal includes:

[0012] Decomposing the target arc signal by at least one signal decomposition method to obtain arc detection feature quantities corresponding to each signal decomposition method;

[0013] Obtaining the second fault identification accuracy corresponding to the arc detection feature quantity corresponding to each signal decomposition method through the accuracy prediction model;

[0014] The initial features of the target arc signal are obtained according to the signal decomposition method corresponding to the maximum second fault identification accuracy.

[0015] In one embodiment, the signal decomposition method is wavelet basis decomposition, and the target arc signal is decomposed to obtain arc detection feature quantities corresponding to each signal decomposition method, including:

[0016] Perform multi-layer resolution decomposition on the target arc signal according to each signal decomposition method to obtain candidate arc signals corresponding to each signal decomposition method;

[0017] Taking the preset sampling point as a time window, the arc detection feature quantity of each time window in the candidate arc signal corresponding to each signal decomposition method is calculated;

[0018] The arc detection feature quantity corresponding to each signal decomposition method is determined according to the arc detection feature quantity of each time window in the candidate arc signal corresponding to each signal decomposition method.

[0019] In one embodiment, determining at least one candidate feature combination from the multiple types of features according to their importance scores includes:

[0020] Obtain importance scores for multiple types of features;

[0021] In descending order of the importance scores of multiple types of features, the first feature among the multiple types of features is used as a candidate feature combination, and the next adjacent feature is added in sequence on the basis of the first feature to obtain each candidate feature combination; wherein, each time the next adjacent feature is added, it corresponds to a candidate feature combination.

[0022] In one embodiment, the method further comprises:

[0023] According to the importance scores of the multiple types of features, features that do not meet the preset combination conditions are eliminated from the importance scores of the multiple types of features.

[0024] In one embodiment, the process of constructing the accuracy prediction model includes:

[0025] Acquire multiple historical arc signal features;

[0026] Multiple historical arc signal features are input into the initial accuracy prediction model to obtain a fault identification accuracy test value, until the fault identification accuracy test value and the preset fault identification accuracy standard value meet the preset iterative convergence condition, and the accuracy prediction model is obtained.

[0027] In one embodiment, determining a target feature combination based on the first fault identification accuracy of each candidate feature combination includes:

[0028] Obtaining a stable interval of the first fault identification accuracy of each candidate feature combination according to the first fault identification accuracy of each candidate feature combination;

[0029] The candidate feature combination corresponding to the first first fault identification accuracy in the stable interval is determined as the target feature combination.

[0030] In a second aspect, an embodiment of the present application provides a feature selection device, the device comprising:

[0031] An acquisition module is used to acquire initial features of a target arc signal; the initial features include multiple types of features;

[0032] A selection module is used to sort the features of the multiple types according to their importance scores and determine at least one candidate feature combination from the multiple types of features;

[0033] An input module, configured to input each candidate feature combination into a preset accuracy prediction model to obtain a first fault identification accuracy of each candidate feature combination;

[0034] The determination module is used to determine the target feature combination according to the first fault identification accuracy of each candidate feature combination.

[0035] In a third aspect, an embodiment of the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method provided in any embodiment of the first aspect when executing the computer program.

[0036] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method provided in any embodiment of the first aspect above.

[0037] The embodiments of the present application provide a feature selection method, apparatus, computer equipment, and storage medium. The method obtains the initial features of a target arc signal, sorts the initial features according to their importance scores, and determines at least one candidate feature combination from the initial features. Each candidate feature combination is input into a preset accuracy prediction model to obtain a first fault identification accuracy of each candidate feature combination. Finally, a target feature combination is determined based on the first fault identification accuracy of each candidate feature combination. In this method, since the initial features include multiple types of features, the initial features can fully reflect the characteristics of the data. The candidate feature combinations are determined by sorting the initial features according to the importance scores. The importance scores are a measure of the importance of the features in the data. Therefore, the candidate feature combinations can be determined more accurately. The candidate feature combinations are input into a preset accuracy prediction model to obtain the first fault identification accuracy of each candidate combination. The target feature combination can be determined based on the obtained first fault identification accuracy. Because the candidate feature combinations are determined based on the importance scores of the features, and the accuracy of each candidate combination is obtained, the target feature combination can be obtained based on the comparison of the accuracy rates. Therefore, this method can quickly and accurately screen out suitable features. The screened out suitable features can achieve the highest accuracy arc fault identification result with the least features. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 FIG. 1 is an application environment diagram of a feature selection method in one embodiment;

[0039] Figure 2 is a flow chart of a feature selection method in one embodiment;

[0040] Figure 3 is a flow chart of a feature selection method in another embodiment;

[0041] Figure 4 A schematic diagram of the results of a feature selection method in one embodiment;

[0042] Figure 5a A schematic diagram of the results of a feature selection method in another embodiment;

[0043] Figure 5b A schematic diagram of the results of a feature selection method in another embodiment;

[0044] Figure 6 is a flow chart of a feature selection method in another embodiment;

[0045] Figure 7 A schematic diagram of the results of a feature selection method in another embodiment;

[0046] Figure 8is a flow chart of a feature selection method in another embodiment;

[0047] Figure 9 A schematic diagram of the results of a feature selection method in another embodiment;

[0048] Figure 10 is a flow chart of a feature selection method in another embodiment;

[0049] Figure 11 is a flow chart of a feature selection method in another embodiment;

[0050] Figure 12 A schematic diagram of the results of a feature selection method in another embodiment;

[0051] Figure 13 is a flow chart of a feature selection method in another embodiment;

[0052] Figure 14 is a structural block diagram of a feature selection device in one embodiment;

[0053] Figure 15 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0055] The feature selection method provided in this application can be applied to computer devices, which can be devices in any field, such as power equipment, or various personal computers, laptops, tablet computers, wearable devices, etc. The embodiment of this application does not limit the type of computer devices. Figure 1 As shown, a schematic diagram of the internal structure of a computer device is provided. Figure 1 The processor in the system is used to provide computing and control capabilities. The memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database is used to store data related to the feature selection process. The network interface is used to communicate with other external devices via a network connection. When executed by the processor, the computer program implements a feature selection method.

[0056] When researching a problem, it's important to obtain as many features as possible. Modeling based on only three or two features is ineffective and lacks interpretability. Having more features provides more information to accurately describe the problem, making the research results more convincing. However, with the continuous emergence of large-scale data processing problems such as genetic engineering, text classification, and image retrieval, large-scale data feature selection poses a severe challenge to existing feature selection algorithms, and feature selection has attracted widespread interest among scholars in the field of machine learning.

[0057] Feature selection usually selects a feature subset that has strong correlation with the category and weak correlation between features. The specific correlation selection algorithm reflects this by defining an appropriate subset evaluation function.

[0058] As a preprocessing step, feature selection plays a crucial role in data mining, pattern recognition, and machine learning. It can reduce the complexity of problems and improve the predictive accuracy, robustness, and interpretability of learning algorithms. While significant progress has been made in feature selection, it remains a complex problem and requires considerable research.

[0059] In one approach, a feature selection algorithm is studied using a Filter algorithm, which has a fast running speed but a low accuracy.

[0060] In another approach, a feature selection method is studied using a wrapper algorithm, which has low efficiency but high accuracy.

[0061] Based on this, the embodiments of the present application provide a feature selection method, apparatus, computer device and storage medium, which can quickly and accurately screen out suitable features, thereby achieving the highest accuracy with the least features.

[0062] The following will describe in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems through embodiments and in conjunction with the accompanying drawings. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. It should be noted that the feature selection method provided by the present application, the execution subject of each embodiment can be a computer device, wherein the execution subject can also be a feature selection device, which can be implemented as part or all of the processor through software, hardware, or a combination of software and hardware.

[0063] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0064] In one embodiment, Figure 2 As shown, a feature selection method is provided. This embodiment involves determining at least one candidate feature combination based on the importance scores of multiple types of features, obtaining the first fault identification accuracy of each candidate feature combination, and determining the target feature combination based on the first fault identification accuracy of each candidate feature combination. This embodiment includes the following steps:

[0065] S201, obtaining initial features of a target arc signal; the initial features include multiple types of features.

[0066] Arcing is a gas discharge phenomenon, a momentary spark produced by current passing through certain insulating media (such as air). Arcing is a high-temperature, highly conductive, free gas that not only severely damages contacts but also prolongs the time it takes to disconnect the circuit. Therefore, if overloaded or poorly connected for a long time, the insulation of the transmission line can easily age, reducing its effectiveness or causing insulation damage, which can lead to arc faults. Therefore, timely detection of arcing is essential.

[0067] In one embodiment, the arc signal may be obtained by installing some parameter detection sensors at a part where an arc may occur, and using the sensors to obtain the arc signal.

[0068] In another embodiment, the arc signal can be obtained by monitoring the electromagnetic radiation, heat and noise during arc discharge using an infrared receiver based on the phenomena of electromagnetic radiation, light, sound and heat during arc discharge, thereby obtaining the arc signal.

[0069] Optionally, the target arc signals are some arc signals required in the embodiments of the present application, and can be directly obtained from a historical database.

[0070] The characteristics of an arc signal can be considered from two perspectives: its time domain characteristics and its frequency domain characteristics. The time domain characteristics of an arc can be characterized by parameters such as the root mean square (RMS), average value, peak value, and standard deviation of the arc signal. The frequency domain characteristics of an arc can be characterized by Fourier transform, which detects changes in the spectrum before and after the arc occurs. Based on these changes in the spectrum, the characteristic frequency bands and frequency domain characteristics of the arc can be determined. Therefore, the initial characteristics of the target arc signal can optionally be the RMS, average value, peak value, standard deviation, and frequency domain characteristics.

[0071] In one embodiment, the initial features of the target arc signal may be obtained by using a pre-trained neural network model, taking the target arc signal as the input of the neural network model, and finally outputting the initial features of the target arc signal through training of the neural network model.

[0072] In another embodiment, the method of obtaining the initial characteristics of the target arc signal may also be to directly input the target arc signal into the calculation formula through a preset calculation method based on the obtained target arc signal to directly obtain the initial characteristics of the target arc signal.

[0073] In another embodiment, the initial characteristics of the target arc signal may be obtained by directly obtaining the initial characteristics of the target arc signal from a database that pre-stores the characteristics of the arc signal.

[0074] S202 , sorting the features of the multiple types according to their importance scores, and determining at least one candidate feature combination from the multiple types of features.

[0075] The importance score measures the value of a feature in the data. The higher the importance score, the more relevant the feature is to the data as a whole.

[0076] The feature combination is determined by the importance scores of the features. Specifically, the importance scores of multiple types of features are sorted, and then a feature combination is determined from the multiple types of features, and the determined feature combination is used as a candidate feature combination; wherein there is at least one candidate feature combination.

[0077] In one embodiment, importance scores of multiple types of features are sorted, and the sorting method may be to sort the importance scores of multiple types of features from large to small.

[0078] In another embodiment, the importance scores of multiple types of features are sorted, and the sorting method may also be to sort the importance scores of multiple types of features in an ascending order.

[0079] In another embodiment, the importance scores of multiple types of features are sorted, and the sorting method can also be based on a preset neural network model. By taking the importance scores of multiple types of features as the input of the preset neural network model, the preset neural network model is trained, and finally the sorted importance scores and corresponding features are output.

[0080] In one embodiment, the method for determining at least one candidate feature combination can be based on a preset feature combination algorithm. Specifically, the importance scores of the sorted multiple types of features are used as input to the preset feature combination algorithm, and at least one candidate feature combination is directly output by running the algorithm.

[0081] S203: Input each candidate feature combination into a preset accuracy prediction model to obtain a first fault identification accuracy of each candidate feature combination.

[0082] The preset accuracy prediction model is a prediction model obtained by training with training set data. The accuracy of existing data can be predicted by the preset accuracy prediction model. The training set data can be a variety of data sets and must be unbiased, comprehensive, and balanced.

[0083] The fault identification accuracy is the accuracy of each candidate combination obtained by training the preset accuracy prediction model.

[0084] In one embodiment, the preset accuracy prediction model can be obtained through a machine learning algorithm, and the preset accuracy prediction model is finally obtained by training the training set through the machine learning algorithm.

[0085] In another embodiment, the preset accuracy prediction model can also be obtained through a neural network algorithm. First, the inceptionV4 network and optimizer are selected, and then the training set is iteratively trained to finally obtain the preset accuracy prediction model.

[0086] In one embodiment, each candidate feature combination obtained in the above embodiment is used as an input of a preset accuracy prediction model, and then the preset accuracy prediction model is trained to obtain a first fault identification accuracy of each candidate feature combination.

[0087] In another embodiment, a state prediction model is first trained, and each candidate feature combination obtained in the above embodiment is used as the input of the state prediction model to obtain the predicted state corresponding to each candidate feature combination. Based on the obtained predicted state, the first fault identification accuracy of each candidate feature combination is further obtained.

[0088] S204: Determine a target feature combination based on the first fault identification accuracy of each candidate feature combination.

[0089] The target feature combination is the best feature subset obtained from multiple features. The target feature combination can achieve the highest accuracy while using the least features.

[0090] In one embodiment, a target feature combination is determined based on the first fault identification accuracy of each candidate feature combination. The determination method can be to use a pre-trained neural network model, use the first fault identification accuracy of each candidate feature combination as the input of the neural network model, and output the target feature combination by training the neural network model.

[0091] In another embodiment, a target feature combination is determined based on the first fault identification accuracy of each candidate feature combination. The determination method can also be to obtain the candidate feature combination corresponding to the highest accuracy through comparison based on the first fault accuracy of each candidate feature combination, and determine the candidate feature combination corresponding to the highest accuracy as the target feature combination.

[0092] In another embodiment, a target feature combination is determined based on the first fault identification accuracy of each candidate feature combination. The determination method can also be to obtain the candidate feature combination corresponding to the minimum accuracy through comparison based on the first fault accuracy of each candidate feature combination, and determine the candidate feature combination corresponding to the minimum accuracy as the target feature combination.

[0093] The feature selection method provided by the embodiment of the present application obtains the initial features of the target arc signal, sorts the initial features according to their importance scores, determines at least one candidate feature combination from the initial features, inputs each candidate feature combination into a preset accuracy prediction model, obtains a first fault identification accuracy of each candidate feature combination, and finally determines a target feature combination based on the first fault identification accuracy of each candidate feature combination. In this method, because the initial features include multiple types of features, the initial features can fully reflect the characteristics of the data. By sorting the initial features according to their importance scores, the candidate feature combinations are determined. The importance score is a measure of the importance of the features in the data. Therefore, the candidate feature combinations can be determined more accurately. The candidate feature combinations are input into the preset accuracy prediction model to obtain a first fault identification accuracy of each candidate combination. The target feature combination can be determined based on the obtained first fault identification accuracy. Because the candidate feature combinations are determined based on the feature importance scores, and the accuracy of each candidate combination is obtained, the target feature combination can be obtained based on the comparison of the accuracy rates. Therefore, the method can quickly and accurately screen out suitable features. The screened suitable features can achieve the highest accuracy arc fault identification result with the least number of features.

[0094] Based on the above embodiments, in one embodiment, Figure 3 As shown, obtaining the initial characteristics of the target arc signal includes the following steps:

[0095] S301 , performing signal decomposition on a target arc signal by using at least one signal decomposition method to obtain arc detection feature quantities corresponding to each signal decomposition method.

[0096] In order to facilitate the analysis of complex signals, the signals can be decomposed. For example, a signal with a non-zero average value can be decomposed into a DC component and an AC component; or an arbitrary signal can be decomposed into an even component and an odd component.

[0097] Signal decomposition methods include Fourier transform and wavelet transform. Among them, Fourier transform transforms the negative infinite time domain signal into the frequency domain for analysis, which reflects the spectrum of the original signal in the entire time; wavelet transform is to transform the basic wavelet or the target wavelet function, which is time-shifted and, under different scale factors, is inner-producted with the given square-integrable time domain original signal. Wavelet transform has the ability to reflect the local characteristics of the signal in both time and frequency domains and has the characteristics of multi-resolution analysis.

[0098] Optionally, the target arc signal may be a 500 kHz arc signal.

[0099] In one embodiment, the target arc signal is decomposed by a signal decomposition method. Specifically, the target arc signal can be decomposed by a Fourier transform method.

[0100] In another embodiment, the target arc signal is decomposed by a signal decomposition method. Specifically, the target arc signal can be decomposed by a wavelet transform method.

[0101] Optionally, the arc detection feature quantity corresponding to each signal decomposition method is obtained. The method for obtaining it can be to extract the average value and peak value of the target arc signal after signal decomposition, and calculate the time domain feature quantity of the target arc signal after signal decomposition based on the average value and peak value, so as to obtain the arc detection feature quantity corresponding to each signal decomposition method; the method for obtaining it can also be based on the wavelet coefficients of the target arc signal after signal decomposition, and extract the feature quantity from the obtained wavelet coefficients as the frequency domain feature quantity, that is, the arc detection feature quantity corresponding to each signal decomposition method.

[0102] S302 , obtaining a second fault identification accuracy corresponding to the arc detection feature quantity corresponding to each signal decomposition method through an accuracy prediction model.

[0103] In one embodiment, a method for obtaining the second fault identification accuracy corresponding to the arc detection feature quantity corresponding to each signal decomposition method may be to use the arc detection feature quantity corresponding to each signal decomposition method as the input of the accuracy prediction model according to the accuracy prediction model, and by running the accuracy prediction model, finally directly output the second fault identification accuracy corresponding to the arc detection feature quantity corresponding to each signal decomposition method.

[0104] One embodiment, such as Figure 4 As shown, Figure 4The arc detection feature corresponding to each signal decomposition method is obtained by selecting different wavelet bases for wavelet decomposition and performing signal decomposition on the target arc signal. Then, the second fault identification accuracy corresponding to the arc detection feature corresponding to each signal decomposition method is obtained based on the accuracy prediction model and the arc detection feature corresponding to each signal decomposition method. Figure 4 That is, the corresponding second fault identification accuracy rate is obtained according to different wavelet bases. By calculating the second fault identification accuracy rate according to different wavelet bases, the importance of the influence of different wavelet base features on the final detection accuracy rate can be obtained. Figure 5a As shown in the figure, it is the actual arc state of the arc detection feature quantity. The number 1 represents the fault arc state, and the number 0 represents the normal arc state. The prediction results of the arc detection feature quantity corresponding to the signal decomposition method of the Rbio3.1 wavelet basis are as follows: Figure 5b As shown, the second fault identification accuracy is obtained according to the actual arc state and the corresponding prediction result.

[0105] S303 : Acquire initial features of the target arc signal according to a signal decomposition method corresponding to the maximum second fault identification accuracy.

[0106] exist Figure 4 The second fault identification accuracy of the Rbio3.1 wavelet basis is 0.947, and the second fault identification accuracy of the feature detection constructed by other wavelet bases is 0.869. Therefore, the Rbio3.1 wavelet basis is more suitable for the construction of fault arc detection features. Figure 4 It can be seen that the detection features constructed based on the Rbio3.1 wavelet basis can more accurately detect the occurrence of fault arcs. The feature importance is positively correlated with the final algorithm detection accuracy. The higher the feature importance, the higher the corresponding accuracy. Therefore, the Rbio3.1 wavelet basis is the most appropriate.

[0107] Optionally, the initial features of the target arc signal are obtained according to the signal decomposition method corresponding to the maximum second fault identification accuracy, such as Figure 4 As shown in the figure, the signal decomposition method corresponding to the maximum second fault identification accuracy is to use the Rbio3.1 wavelet basis for 6-layer wavelet decomposition. Therefore, the target arc signal is decomposed into 6 layers of wavelet using the Rbio3.1 wavelet basis, and the decomposed target arc signal in the frequency band of 23.4-31.25kHz is selected. Then, the average value, maximum value, minimum value and sheath of the wavelet coefficients in the time window composed of each 8000 sampling points are calculated, and the average value, maximum value, minimum value and sheath of each time window are used as the initial features of the target arc signal.

[0108] Among them, the method of calculating the initial characteristics of the target arc signal can be based on a preset neural network model, using the signal decomposition method corresponding to the maximum second fault identification accuracy and the target arc signal as the input of the neural network model, and through the training of the neural network model, finally directly outputting the initial characteristics of the target arc signal.

[0109] The feature selection method provided in an embodiment of the present application first decomposes the target arc signal using at least one signal decomposition method to obtain arc detection feature quantities corresponding to each signal decomposition method. Then, using an accuracy prediction model, the second fault identification accuracy corresponding to the arc detection feature quantities corresponding to each signal decomposition method is obtained. Finally, the initial features of the target arc signal are obtained based on the signal decomposition method corresponding to the highest second fault identification accuracy. In this method, arc detection feature quantities corresponding to each signal decomposition method are obtained according to different signal decomposition methods, and then the corresponding second fault identification accuracy is calculated based on the feature quantities to obtain the initial features of the target arc signal. This improves the accuracy of fault detection and provides a basis for quickly and accurately screening suitable feature combinations.

[0110] In the above embodiment, the target arc signal is decomposed to obtain arc detection feature quantities corresponding to each signal decomposition method. This is described in detail below. In one embodiment, Figure 6 As shown, the signal decomposition method is wavelet basis decomposition, and the target arc signal is decomposed to obtain the arc detection feature quantity corresponding to each signal decomposition method, including the following steps:

[0111] S601 , performing multi-layer resolution decomposition on a target arc signal according to various signal decomposition methods to obtain candidate arc signals corresponding to various signal decomposition methods.

[0112] In one embodiment, the signal decomposition method can adopt the wavelet decomposition method. In the wavelet transform, different wavelet bases can be selected. Therefore, different wavelet basis functions can be used to decompose the target arc signal. The wavelet basis functions can include: Rbio3.1, Rbio1.1, Rbio4.4, Rbio3.5, Bior1.1, Bior3.3, Db1, DB2, Sym2, etc.

[0113] In one embodiment, a multi-layer resolution decomposition is performed on the target arc signal. Rbio3.1 can be used as a wavelet basis to perform a 6-layer wavelet decomposition on the target arc signal to obtain a decomposed target arc signal. Then, a signal in the 23.4-31.25kHz frequency band can be selected from the decomposed target arc signal as a candidate arc signal. In actual applications, the embodiment of the present application does not impose any limitation on how many frequency bands are selected.

[0114] In another embodiment, the target arc signal is subjected to multi-layer resolution decomposition. Bior3.1 can also be used as a wavelet basis to perform a 5-layer wavelet decomposition on the current signal to obtain the decomposed target arc signal, and then the decomposed target arc signal can be directly used as a candidate arc signal.

[0115] Optionally, the target arc signal is subjected to multi-layer resolution decomposition. In practical applications, several layers of resolution decomposition are specifically selected, which is not limited in the embodiments of the present application.

[0116] S602 , taking a preset sampling point as a time window, and calculating arc detection feature quantities of each time window in the candidate arc signals corresponding to each signal decomposition method.

[0117] The arc detection characteristic quantity is a characteristic quantity of the target arc obtained by calculating the target arc signal after signal decomposition, and can represent the state of the target arc signal.

[0118] In one embodiment, the candidate arc signal obtained in the above embodiment is used to form an analysis time window according to preset sampling points, and then the characteristic quantity of the signal in each time window is calculated. The characteristic quantity corresponding to each time window in the candidate arc signal is the arc detection characteristic quantity corresponding to the signal decomposition method; the method for calculating the characteristic quantity of the signal in each time window can be to square the sampling points in each time window, and then the sum of each square term is the characteristic quantity corresponding to each time window.

[0119] Optionally, the preset sampling points may be 8000 sampling points, but in practical applications, the embodiment of the present application is not limited to this.

[0120] In one embodiment, a method for calculating the arc detection feature quantity of each time window in the candidate arc signal corresponding to each signal decomposition method can be obtained through a pre-configured program. Specifically, each time window in the candidate arc signal corresponding to each signal decomposition method is used as input, and by calling the pre-configured program, the arc detection feature quantity of each time window in the candidate arc signal corresponding to each signal decomposition method is finally obtained.

[0121] S603 , determining arc detection feature quantities corresponding to each signal decomposition method according to arc detection feature quantities of each time window in the candidate arc signal corresponding to each signal decomposition method.

[0122] In one embodiment, the arc detection feature quantity corresponding to each signal decomposition method is determined based on the arc detection feature quantity of each time window in the candidate arc signal corresponding to each signal decomposition method. The determination method can be to directly determine the arc detection feature quantity of each time window in the candidate arc signal corresponding to each signal decomposition method as the arc detection feature quantity corresponding to each signal decomposition method; for example, if the candidate arc signal corresponding to a signal decomposition method has 3 time windows, and the arc detection feature quantities of each time window are a, b and c respectively, then the arc detection feature quantities corresponding to the signal decomposition method are a, b and c.

[0123] Another way is to determine the arc detection feature quantity corresponding to each signal decomposition method based on the arc detection feature quantity of each time window in the candidate arc signal corresponding to each signal decomposition method. The determination method can also be calculated through a calculation function. Specifically, the arc detection feature quantity of each time window in the candidate arc signal corresponding to each signal decomposition method is substituted into the calculation function, and the final function result is the arc detection feature quantity corresponding to each signal decomposition method; for example, if the candidate arc signal corresponding to a signal decomposition method has 3 time windows, the arc detection feature quantities of each time window are a, b and c respectively, and the arc detection feature quantities of each time window are substituted into the calculation function respectively, the results obtained are d, e and f respectively, then the arc detection feature quantities corresponding to the signal decomposition method are d, e and f.

[0124] In one embodiment, a 6-layer wavelet decomposition is performed on the target arc signal according to the Rbio.3 wavelet basis, and the decomposed target arc signal of 23.4-31.25kHz is selected. Every 8000 sampling points are regarded as an analysis time window, and the square term of each sampling point in the time window is calculated. The sum of the square terms of each sampling point is regarded as the arc detection feature of each time window, such as Figure 7 As shown, Figure 7 is the arc detection feature obtained based on the target arc signal.

[0125] The feature selection method provided in the embodiments of the present application performs multi-layer resolution decomposition on the target arc signal according to various signal decomposition methods to obtain candidate arc signals corresponding to each signal decomposition method. The method uses a preset sampling point as a time window to calculate arc detection feature quantities for each time window in the candidate arc signals corresponding to each signal decomposition method. Based on the arc detection feature quantities for each time window in the candidate arc signals corresponding to each signal decomposition method, the arc detection feature quantities corresponding to each signal decomposition method are determined. This method can quickly and accurately select appropriate features, and the selected feature combination can achieve the highest accuracy arc fault identification results using the least number of features.

[0126] Based on the above embodiments, in one embodiment, Figure 8 As shown, sorting the features of multiple types according to their importance scores and determining at least one candidate feature combination from the features of multiple types includes the following steps:

[0127] S801, obtaining importance scores of multiple types of features.

[0128] In one embodiment, importance scores of multiple types of features are obtained by calculating the importance scores of multiple types of features according to a random forest method. Specifically, the importance of the features is measured by calculating the purity of the multiple types of features through the Gini index based on the multiple types of features, thereby obtaining the importance scores of the multiple types of features.

[0129] In another embodiment, importance scores of multiple types of features are obtained. The method for obtaining the importance scores of multiple types of features can also be to calculate the importance scores of multiple types of features based on the eXtreme Gradient Boosting (XGBoost) algorithm. Specifically, a decision tree is set according to the multiple types of features, and the importance scores of multiple types of features are calculated through the decision tree.

[0130] S802, in descending order of importance scores of the multiple types of features, taking the first feature of the multiple types of features as a candidate feature combination, and sequentially adding the next adjacent features to the first feature to obtain candidate feature combinations.

[0131] Among them, each adjacent feature is added once and corresponds to a candidate feature combination.

[0132] The candidate feature combination is obtained by combining multiple types of features.

[0133] In one embodiment, the importance scores of multiple types of features are arranged in descending order, the first feature of the multiple types of features is used as a candidate feature combination, and the next adjacent features are added in sequence on the basis of the first candidate combination to obtain each candidate feature combination; for example, the multiple types of features are: entropy, variance, standard deviation, mean square error, skewness, average value, maximum value, minimum value and sheath, and the importance scores of the multiple types of features are arranged in descending order as: variance, minimum value, maximum value, average value, skewness, sheath, mean square error, standard deviation and entropy, such as Figure 9 As shown, Figure 9The importance scores of various types of features are sorted in descending order, and the variance is taken as the first candidate feature. Then the feature of the minimum value is added in sequence, and the variance and minimum value are taken as the second candidate feature combination. In sequence, the variance, minimum value and maximum value are taken as the third candidate feature combination, the variance, minimum value, maximum value and average value are taken as the fourth candidate feature combination, the variance, minimum value, maximum value, average value and skewness are taken as the fifth candidate feature combination, the variance, minimum value, maximum value, average value, skewness and sheath are taken as the sixth candidate feature combination, the variance, minimum value, maximum value, average value, skewness, sheath and mean square error are taken as the seventh candidate feature combination, the variance, minimum value, maximum value, average value, skewness, sheath, mean square error and standard deviation are taken as the eighth candidate feature combination, and the variance, minimum value, maximum value, average value, skewness, sheath, mean square error, standard deviation and entropy are taken as the ninth candidate feature combination, thereby obtaining the candidate feature combinations.

[0134] The feature selection method provided in the embodiment of the present application obtains the importance scores of multiple types of features, and uses the first feature of the multiple types of features as a candidate feature combination in descending order of the importance scores of the multiple types of features, and sequentially adds the next adjacent feature to the first feature to obtain each candidate feature combination; wherein each addition of the next adjacent feature corresponds to a candidate feature combination. In this method, the importance scores of the multiple types of features are first obtained, and then candidate feature combinations are generated in sequence according to the ranking of the importance scores from large to small. Since the importance scores have a high correlation with the overall data, this method can quickly and accurately screen out appropriate features, achieving the highest accuracy arc fault identification result with the least features.

[0135] After obtaining the importance scores of multiple types of features, the importance scores of the multiple types of features are sorted from large to small, and then the candidate feature combinations are obtained. However, before obtaining the candidate feature combinations, there is an operation of eliminating features with lower importance scores. In response to this situation, an embodiment is described below. The embodiment includes: according to the importance scores of multiple types of features, features that do not meet the preset combination conditions are eliminated from the importance scores of the multiple types of features.

[0136] In one embodiment, the preset combination condition may be to set a fixed value, compare the importance scores of multiple types of features with the preset value, and eliminate features with importance scores less than the preset value. For example, multiple types of features include maximum value, minimum value, average value, and skewness, and the importance scores corresponding to the maximum value, minimum value, average value, and skewness are 200, 400, 300, and 350, respectively. The preset combination condition is greater than 280, so the maximum value feature corresponding to the importance score of 200 will be eliminated.

[0137] In another embodiment, the preset combination condition can also be to set a ratio, sort the importance scores of multiple types of features, and then eliminate the features of the preset ratio. When eliminating features, they are eliminated from small to large, and the features with low importance scores are eliminated first. For example, multiple types of features include maximum value, minimum value, average value and skewness, and the importance scores corresponding to the maximum value, minimum value, average value and skewness are 200, 400, 300 and 350 respectively. The preset combination condition is that the preset ratio is 25%, then 1 of the 4 features will be eliminated, and the importance score corresponding to the maximum value is the smallest, so the maximum value feature will be eliminated.

[0138] Based on the above embodiment, each candidate feature combination is input into the preset accuracy prediction model to obtain the first fault identification accuracy of each candidate feature combination and the second fault identification accuracy corresponding to the arc detection feature quantity corresponding to each signal decomposition method is obtained through the accuracy prediction model. The construction process of the accuracy model is described in detail through an embodiment. In one embodiment, Figure 10 As shown in Figure 2, the construction process of the accuracy prediction model includes the following steps:

[0139] S1001, obtaining multiple historical arc signal features.

[0140] The historical arc signal characteristics include various characteristic expressions of the arc signal, including arc signal characteristics obtained from the time domain and frequency domain. The arc states of the arc signal characteristics include two types: fault and normal. Therefore, the historical arc signal characteristics include fault historical arc signal characteristics and normal historical arc signal characteristics.

[0141] In one embodiment, a method for obtaining multiple historical arc signal features may be to send a data request message to a historical database, and then directly obtain the historical arc signal features from the historical database.

[0142] In another embodiment, a method for obtaining multiple historical arc signal characteristics can also be to obtain them from a data acquisition instrument. Specifically, after the data acquisition instrument collects arc data, a data file is generated, where the data file includes arc signal characteristics. Obtaining multiple historical arc signal characteristics means obtaining them from the data file in the data acquisition instrument.

[0143] S1002, input multiple historical arc signal features into the initial accuracy prediction model to obtain a fault identification accuracy test value, until the fault identification accuracy test value and the preset fault identification accuracy standard value meet the preset iterative convergence condition, and obtain the accuracy prediction model.

[0144] The initial accuracy prediction model is an initial accuracy prediction model and can be regarded as a semi-finished accuracy prediction model.

[0145] Multiple historical arc signal features are input into the initial accuracy prediction model to obtain a test value of the fault identification accuracy. The test value of the fault identification accuracy is compared with a preset standard value of the fault identification accuracy to determine whether the iteration condition is met. If the iteration condition is not met, the accuracy prediction model is continued to be iterated. If the preset iteration condition is met, the accuracy prediction model is obtained.

[0146] In one embodiment, multiple historical arc signal features are input into the XGBoost algorithm for training to obtain a test value of the fault identification accuracy. The test value of the fault identification accuracy is compared with a preset standard value of the fault identification accuracy. If the iteration condition is met, an accuracy prediction model is obtained. Otherwise, the iteration is continued until the preset iteration condition is met.

[0147] Among them, optionally, the preset iteration condition can be that the test value of the fault identification accuracy is consistent with the preset standard value of the fault identification accuracy; the preset iteration condition can also be that the test value of the fault identification accuracy and the preset standard value of the fault identification accuracy have a very small interval error.

[0148] The feature selection method provided in an embodiment of the present application obtains multiple historical arc signal features, inputs the multiple historical arc signal features into an initial accuracy prediction model, and obtains a fault identification accuracy test value. This method continues until the fault identification accuracy test value and a preset fault identification accuracy standard value meet a preset iterative convergence condition, thereby obtaining an accuracy prediction model. In this method, the accuracy prediction model is obtained by training multiple historical arc signal features. The accuracy prediction model can accurately predict the fault accuracy of the arc signal features. Therefore, the model can quickly and accurately screen suitable features, achieving the highest accuracy arc fault identification result with the least number of features.

[0149] Based on the above embodiments, in one embodiment, Figure 11 As shown, according to the first fault identification accuracy of each candidate feature combination, determining the target feature combination includes the following steps:

[0150] S1101 , obtaining a stable interval of the first fault discrimination accuracy of each candidate feature combination according to the first fault discrimination accuracy of each candidate feature combination.

[0151] In one embodiment, a method for obtaining the stable interval of the first fault identification accuracy of each candidate feature combination may be obtained through a neural network model, where the first fault identification accuracy of each candidate feature combination is used as the input of the neural network model. Through the training of the neural network model, the neural network model will eventually directly output the candidate feature combination corresponding to the stable interval of the first fault identification accuracy.

[0152] The stable interval of the first fault identification accuracy of each candidate feature combination indicates the candidate feature combination interval where the first fault identification accuracy of each candidate feature combination no longer changes and tends to be stable. Figure 12 As shown, Figure 12 is the first fault discrimination accuracy obtained by combining each candidate feature, which means that the first fault discrimination accuracy of variance is 0.9965, the first fault discrimination accuracy of variance and minimum value is 0.997, the first fault discrimination accuracy of variance, minimum value and maximum value is 0.9986, the first fault discrimination accuracy of variance, minimum value, maximum value and average value is 0.9989, the first fault discrimination accuracy of variance, minimum value, maximum value, average value and skewness is 0.9992, the first fault discrimination accuracy of variance, minimum value, maximum value, average value, skewness and sheath is 0.9992, the first fault discrimination accuracy of variance, minimum value, maximum value, average value, skewness, sheath and mean square error is 0.9992, and the first fault discrimination accuracy of variance, minimum value, maximum value, average value, skewness, sheath and mean square error is 0.9992. The first fault identification accuracy of variance, minimum, maximum, average, skewness, sheath, mean square error and standard deviation is 0.9992, and the first fault identification accuracy of variance, minimum, maximum, average, skewness, sheath, mean square error, standard deviation and entropy is 0.9992; it can be seen that the candidate feature combinations corresponding to the stable interval of the first fault identification accuracy are: variance, minimum, maximum, average and skewness, variance, minimum, maximum, average, skewness and sheath, variance, minimum, maximum, average, skewness, sheath and mean square error, variance, minimum, maximum, average, skewness, sheath, mean square error and standard deviation, variance, minimum, maximum, average, skewness, sheath, mean square error, standard deviation and entropy.

[0153] S1102: Determine the candidate feature combination corresponding to the first first fault identification accuracy rate in the stable interval as the target feature combination.

[0154] Please continue to see Figure 12 , Figure 12 It can be seen that the candidate feature combination corresponding to the first fault identification accuracy in the stable interval is variance, minimum value, maximum value, average value and skewness, so variance, minimum value, maximum value, average value and skewness can be determined as the target feature combination.

[0155] The present application provides a feature selection method that obtains the stable interval of the first fault discrimination accuracy of each candidate feature combination based on the first fault discrimination accuracy of each candidate feature combination, and determines the candidate feature combination corresponding to the first first fault discrimination accuracy in the stable interval as the target feature combination. In this method, the candidate feature combination corresponding to the first first fault discrimination accuracy in the stable interval is determined as the target feature combination, which can quickly and accurately screen out suitable feature combinations, achieving the highest accuracy arc fault discrimination result using the fewest features.

[0156] like Figure 13 As shown, in one embodiment, a feature selection method is also included, and the embodiment includes:

[0157] S1301, using the XGBoost algorithm to obtain a state classification standard for arc data based on the original arc characteristics and arc state.

[0158] S1302, obtaining a DC fault arc signal and its corresponding state, selecting different wavelet bases, performing 6-layer wavelet decomposition on the DC fault arc signal, and then selecting a current signal m1 with a frequency range of 23.4-31.25 kHz;

[0159] Among them, the wavelet bases include: Rbio3.1, Rbio1.1, Rbio4.4, Rbio3.5, Bior1.1, Bior3.3, Db1, DB2, and Sym2.

[0160] S1303: Consider each of the 8,000 sampling points in m1 as an analysis time window, square the sampling points in each time window, and then consider the sum of the squares as the fault arc detection feature value of each analysis time window to obtain the feature value pool of m1 under each wavelet basis.

[0161] S1304: Based on the feature pool, prediction is performed using the state classification standard of the arc data to obtain the predicted state corresponding to each feature under each wavelet basis.

[0162] S1305 , obtaining the detection accuracy rate under each wavelet basis according to the predicted state corresponding to each feature under each wavelet basis and the actual state corresponding to each feature under each wavelet basis.

[0163] S1306: Using the wavelet basis corresponding to the maximum detection accuracy as the wavelet basis for constructing the arc fault detection feature, the DC arc fault signal is subjected to a 6-layer wavelet decomposition. Then, a current signal with a frequency range of 23.4-31.25 kHz is selected, and 8000 sampling points are used as a time window. The feature quantity corresponding to the wavelet coefficient in each time window is calculated to obtain a feature quantity pool.

[0164] The feature quantities include entropy, variance, standard deviation, mean square error, skewness, mean, maximum value, minimum value, and sheath. S1307: Based on the feature quantity pool, the importance score corresponding to each feature is calculated using the XGBoost algorithm.

[0165] S1307 , sorting the importance scores corresponding to the features from large to small, and removing the 10% features with the smallest importance scores.

[0166] S1308 , taking the feature corresponding to the largest importance score as the basic feature, and adding other features one by one to calculate the accuracy of fault identification.

[0167] S1309, when the accuracy rate no longer increases when adding features one by one, the features added after the accuracy rate no longer increases are regarded as redundant features, and the features before the accuracy rate no longer increases are used to determine the final feature combination.

[0168] The implementation principles and technical effects of each step in the feature selection method provided in this embodiment are similar to those in the previous feature selection method embodiments and will not be repeated here.

[0169] It is worth noting that the feature selection method provided in the embodiments of the present application can be applied to a variety of data and has a certain degree of universality and portability.

[0170] It should be understood that, although the various steps in the flowcharts of the above-described embodiments are shown in sequence as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps may be performed in other orders. Moreover, at least a portion of the steps in the flowcharts of the above-described embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily performed at the same time, but may be performed at different times. The execution order of these steps or stages is not necessarily to be performed in sequence, but may be performed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0171] In addition, the present invention also provides a feature selection device. Figure 14As shown, in one embodiment, the feature selection device 1400 includes: an acquisition module 1401, a selection module 1402, an input module 1403 and a determination module 1404, wherein:

[0172] The first acquisition module 1401 is used to acquire the initial features of the target arc signal; the initial features include multiple types of features;

[0173] A selection module 1402 is configured to sort the features of the multiple types according to their importance scores and determine at least one candidate feature combination from the multiple types of features;

[0174] An input module 1403 is configured to input each candidate feature combination into a preset accuracy prediction model to obtain a first fault identification accuracy of each candidate feature combination;

[0175] The determination module 1404 is configured to determine a target feature combination based on the first fault identification accuracy of each candidate feature combination.

[0176] In one embodiment, the first acquisition module 1401 includes:

[0177] a decomposition unit, configured to decompose the target arc signal by using at least one signal decomposition method to obtain arc detection feature quantities corresponding to each signal decomposition method;

[0178] A first acquisition unit is configured to acquire, by using an accuracy prediction model, a second fault identification accuracy corresponding to the arc detection feature quantity corresponding to each signal decomposition method;

[0179] The second acquisition unit is configured to acquire the initial characteristics of the target arc signal according to a signal decomposition method corresponding to the maximum second fault identification accuracy.

[0180] In one embodiment, the decomposition unit comprises:

[0181] a decomposition subunit, configured to perform multi-layer resolution decomposition on the target arc signal according to each signal decomposition method, and obtain candidate arc signals corresponding to each signal decomposition method;

[0182] A calculation subunit, configured to use a preset sampling point as a time window and calculate arc detection feature quantities of each time window in the candidate arc signal corresponding to each signal decomposition method;

[0183] The determination subunit is configured to determine the arc detection feature quantity corresponding to each signal decomposition method according to the arc detection feature quantity of each time window in the candidate arc signal corresponding to each signal decomposition method.

[0184] In one embodiment, the selection module 1402 includes:

[0185] A third acquisition unit is used to obtain importance scores of multiple types of features;

[0186] The combination unit is used to use the first feature of the multiple types of features as a candidate feature combination in descending order of the importance scores of the multiple types of features, and to sequentially add the next adjacent feature on the basis of the first feature to obtain each candidate feature combination; wherein each addition of the next adjacent feature corresponds to a candidate feature combination.

[0187] In one embodiment, the apparatus further comprises:

[0188] The elimination module is used to eliminate features that do not meet the preset combination conditions among the importance scores of multiple types of features according to the importance scores of multiple types of features.

[0189] In one embodiment, a feature selection device is further provided, the device comprising:

[0190] A second acquisition module is used to acquire multiple historical arc signal features;

[0191] The training module is used to input multiple historical arc signal features into the initial accuracy prediction model to obtain a fault identification accuracy test value, until the fault identification accuracy test value and the preset fault identification accuracy standard value meet the preset iterative convergence condition, thereby obtaining an accuracy prediction model.

[0192] In one embodiment, the determining module 1404 includes:

[0193] A fourth acquiring unit, configured to acquire a stable interval of the first fault identification accuracy of each candidate feature combination according to the first fault identification accuracy of each candidate feature combination;

[0194] The determining unit is configured to determine the candidate feature combination corresponding to the first first fault identification accuracy rate in the stable interval as the target feature combination.

[0195] For the specific definition of the feature selection device, please refer to the definition of the feature selection method above, which will not be repeated here. The various modules in the above-mentioned feature selection device can be implemented in whole or in part by software, hardware, or a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0196] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 15As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a feature selection method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad provided on the computer device housing, or an external keyboard, touchpad or mouse.

[0197] Those skilled in the art will understand that Figure 15 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0198] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0199] Acquire initial features of a target arc signal; the initial features include multiple types of features;

[0200] sorting the features of the multiple types according to their importance scores, and determining at least one candidate feature combination from the multiple types of features;

[0201] Inputting each candidate feature combination into a preset accuracy prediction model to obtain the first fault identification accuracy of each candidate feature combination;

[0202] The target feature combination is determined according to the first fault identification accuracy of each candidate feature combination.

[0203] In one embodiment, the processor implements the following steps when executing the computer program:

[0204] Decomposing the target arc signal by at least one signal decomposition method to obtain arc detection feature quantities corresponding to each signal decomposition method;

[0205] Obtaining the second fault identification accuracy corresponding to the arc detection feature quantity corresponding to each signal decomposition method through the accuracy prediction model;

[0206] The initial features of the target arc signal are obtained according to the signal decomposition method corresponding to the maximum second fault identification accuracy.

[0207] In one embodiment, the processor implements the following steps when executing the computer program:

[0208] Perform multi-layer resolution decomposition on the target arc signal according to each signal decomposition method to obtain candidate arc signals corresponding to each signal decomposition method;

[0209] Taking the preset sampling point as a time window, the arc detection feature quantity of each time window in the candidate arc signal corresponding to each signal decomposition method is calculated;

[0210] The arc detection feature quantity corresponding to each signal decomposition method is determined according to the arc detection feature quantity of each time window in the candidate arc signal corresponding to each signal decomposition method.

[0211] In one embodiment, the processor implements the following steps when executing the computer program:

[0212] Obtain importance scores for multiple types of features;

[0213] In descending order of the importance scores of multiple types of features, the first feature among the multiple types of features is used as a candidate feature combination, and the next adjacent feature is added in sequence on the basis of the first feature to obtain each candidate feature combination; wherein, each time the next adjacent feature is added, it corresponds to a candidate feature combination.

[0214] In one embodiment, the processor implements the following steps when executing the computer program:

[0215] According to the importance scores of the multiple types of features, features that do not meet the preset combination conditions are eliminated from the importance scores of the multiple types of features.

[0216] In one embodiment, the processor implements the following steps when executing the computer program:

[0217] Obtain multiple historical arc signal characteristics and corresponding standard arc states;

[0218] Multiple historical arc signal features are input into the initial accuracy prediction model to obtain a fault identification accuracy test value, until the fault identification accuracy test value and the preset fault identification accuracy standard value meet the preset iterative convergence condition, and the accuracy prediction model is obtained.

[0219] In one embodiment, the processor implements the following steps when executing the computer program:

[0220] Obtaining a stable interval of the first fault identification accuracy of each candidate feature combination according to the first fault identification accuracy of each candidate feature combination;

[0221] The candidate feature combination corresponding to the first first fault identification accuracy in the stable interval is determined as the target feature combination.

[0222] The computer device provided in the above embodiment has an implementation principle and technical effects similar to those of the above method embodiment, and will not be described in detail here.

[0223] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0224] Acquire initial features of a target arc signal; the initial features include multiple types of features;

[0225] sorting the features of the multiple types according to their importance scores, and determining at least one candidate feature combination from the multiple types of features;

[0226] Inputting each candidate feature combination into a preset accuracy prediction model to obtain the first fault identification accuracy of each candidate feature combination;

[0227] The target feature combination is determined according to the first fault identification accuracy of each candidate feature combination.

[0228] In one embodiment, when the computer program is executed by a processor, the following steps are implemented:

[0229] Decomposing the target arc signal by at least one signal decomposition method to obtain arc detection feature quantities corresponding to each signal decomposition method;

[0230] Obtaining the second fault identification accuracy corresponding to the arc detection feature quantity corresponding to each signal decomposition method through the accuracy prediction model;

[0231] The initial features of the target arc signal are obtained according to the signal decomposition method corresponding to the maximum second fault identification accuracy.

[0232] In one embodiment, when the computer program is executed by a processor, the following steps are implemented:

[0233] Perform multi-layer resolution decomposition on the target arc signal according to each signal decomposition method to obtain candidate arc signals corresponding to each signal decomposition method;

[0234] Taking the preset sampling point as a time window, the arc detection feature quantity of each time window in the candidate arc signal corresponding to each signal decomposition method is calculated;

[0235] The arc detection feature quantity corresponding to each signal decomposition method is determined according to the arc detection feature quantity of each time window in the candidate arc signal corresponding to each signal decomposition method.

[0236] In one embodiment, when the computer program is executed by a processor, the following steps are implemented:

[0237] Obtain importance scores for multiple types of features;

[0238] In descending order of the importance scores of multiple types of features, the first feature among the multiple types of features is used as a candidate feature combination, and the next adjacent feature is added in sequence on the basis of the first feature to obtain each candidate feature combination; wherein, each time the next adjacent feature is added, it corresponds to a candidate feature combination.

[0239] In one embodiment, when the computer program is executed by a processor, the following steps are implemented:

[0240] According to the importance scores of the multiple types of features, features that do not meet the preset combination conditions are eliminated from the importance scores of the multiple types of features.

[0241] In one embodiment, when the computer program is executed by a processor, the following steps are implemented:

[0242] Obtain multiple historical arc signal characteristics and corresponding standard arc states;

[0243] Multiple historical arc signal features are input into the initial accuracy prediction model to obtain a fault identification accuracy test value, until the fault identification accuracy test value and the preset fault identification accuracy standard value meet the preset iterative convergence condition, and the accuracy prediction model is obtained.

[0244] In one embodiment, when the computer program is executed by a processor, the following steps are implemented:

[0245] Obtaining a stable interval of the first fault identification accuracy of each candidate feature combination according to the first fault identification accuracy of each candidate feature combination;

[0246] The candidate feature combination corresponding to the first first fault identification accuracy in the stable interval is determined as the target feature combination.

[0247] The above embodiment provides a computer-readable storage medium, whose implementation principle and technical effects are similar to those of the above method embodiment, and will not be repeated here.

[0248] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0249] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0250] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A feature selection method, characterized in that: The method comprises: Acquiring initial features of a target arc signal; the initial features include multiple types of features; According to the multiple types of features, the purity of the multiple types of features is calculated by using the Gini index to measure the importance of the features, thereby obtaining importance scores of the multiple types of features; the importance score of each feature represents the relevance of the feature to the integrity of the data; Eliminate a preset proportion of features according to the importance scores of the multiple types of features in ascending order; use the first feature of the multiple types of features after elimination as a candidate feature combination, and sequentially add the next adjacent feature to the first feature to obtain each candidate feature combination; wherein each addition of the next adjacent feature corresponds to a candidate feature combination; Inputting each candidate feature combination into a preset accuracy prediction model to obtain a first fault identification accuracy of each candidate feature combination; According to the changing trend of the first fault identification accuracy of each candidate feature combination, a stable interval consisting of N consecutive candidate feature combinations after the first fault identification accuracy reaches a preset threshold for the first time is obtained; N is a positive integer greater than 1; The candidate feature combination with the least number of features in the stable interval is determined as the target feature combination.

2. The method according to claim 1, characterized in that The obtaining of the initial characteristics of the target arc signal includes: Decomposing the target arc signal by at least one signal decomposition method to obtain arc detection feature quantities corresponding to each signal decomposition method; Obtaining, by means of the accuracy prediction model, a second fault identification accuracy corresponding to the arc detection feature quantity corresponding to each of the signal decomposition modes; The initial features of the target arc signal are obtained according to the signal decomposition method corresponding to the maximum second fault identification accuracy.

3. The method according to claim 2, characterized in that The signal decomposition method is wavelet basis decomposition, and the target arc signal is decomposed to obtain arc detection feature quantities corresponding to each signal decomposition method, including: Performing multi-layer resolution decomposition on the target arc signal according to each of the signal decomposition methods to obtain candidate arc signals corresponding to each of the signal decomposition methods; Taking the preset sampling point as a time window, calculating the arc detection feature value of each time window in the candidate arc signal corresponding to each signal decomposition method; The arc detection feature quantity corresponding to each signal decomposition method is determined according to the arc detection feature quantity of each time window in the candidate arc signal corresponding to each signal decomposition method.

4. The method according to any one of claims 1 to 3, characterized in that The method further comprises: The importance scores of the sorted features of various types are used as input to a preset feature combination algorithm, and at least one candidate feature combination is output by running the algorithm.

5. The method according to any one of claims 1 to 3, characterized in that The construction process of the accuracy prediction model includes: Acquire multiple historical arc signal features; The multiple historical arc signal features are input into the initial accuracy prediction model to obtain a fault identification accuracy test value, until the fault identification accuracy test value and the preset fault identification accuracy standard value meet the preset iterative convergence condition, thereby obtaining the accuracy prediction model.

6. The method according to claim 5, characterized in that The preset iteration condition is that the test value of the fault identification accuracy is consistent with the preset fault identification accuracy.

7. A feature selection device, characterized in that: The device comprises: An acquisition module, configured to acquire initial features of a target arc signal; the initial features include multiple types of features; A selection module is configured to calculate the purity of the multiple types of features using the Gini index to measure the importance of the features, thereby obtaining importance scores for the multiple types of features; the importance score of each feature represents the relevance of the feature to the integrity of the data; and to eliminate a preset proportion of features in ascending order of the importance scores of the multiple types of features; to use the first feature of the multiple types of features after elimination as a candidate feature combination, and to sequentially add subsequent adjacent features to the first feature to obtain each candidate feature combination; wherein each addition of the subsequent adjacent feature corresponds to a candidate feature combination; An obtaining module, configured to input each candidate feature combination into a preset accuracy prediction model to obtain a first fault identification accuracy of each candidate feature combination; A determination module is used to obtain, based on the changing trend of the first fault identification accuracy of each candidate feature combination, a stable interval consisting of N consecutive candidate feature combinations after the first fault identification accuracy first reaches a preset threshold; N is a positive integer greater than 1; and the candidate feature combination with the least number of features in the stable interval is determined as the target feature combination.

8. The device according to claim 7, characterized in that The acquisition module includes: a decomposition unit, configured to perform signal decomposition on the target arc signal by using at least one signal decomposition method, and obtain arc detection feature quantities corresponding to each signal decomposition method; a first acquiring unit, configured to acquire, by means of the accuracy prediction model, a second fault identification accuracy corresponding to the arc detection feature quantity corresponding to each of the signal decomposition modes; The second acquisition unit is configured to acquire the initial features of the target arc signal according to a signal decomposition method corresponding to the maximum second fault identification accuracy.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images