Series fault arc detection method and device based on multi-dimensional feature fusion

Abstract

The invention discloses a series fault arc detection method and device based on multi-dimensional feature fusion. The method comprises the steps of S01, collecting a line current signal and load end voltage data in real time; s02, extracting a wavelet energy entropy, a WRI feature and an arc gap voltage of the line current, and fusing the wavelet energy entropy, the WRI feature and the arc gap voltage to form a multi-dimensional discrimination feature; s03, the extracted multi-dimensional discrimination features are input into a pre-trained series fault arc detection model, whether a series fault arc exists at present or not is recognized, and the series fault arc detection model is a random forest-based model in advance. The method is obtained by training a series fault arc multi-dimensional discrimination feature data set formed by combining the wavelet energy entropy of the line current, the WRI feature and the arc gap voltage in a normal state and a series fault arc state under different load working conditions. The method has the advantages of being simple in implementation method, low in cost, high in detection efficiency and accuracy, low in missed judgment and misjudgment rate and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical line fault arc detection, and in particular to a series fault arc detection method and device based on multi-dimensional feature fusion. Background Art

[0002] At present, electrical lines are deeply embedded in users' living areas. Due to the complex wiring and the susceptibility of electrical lines to user behavior, various types of safety hazards will arise, among which fault arcs will pose a great threat to the safety of power users. For example, insulation damage and electrical aging will cause fault arcs, which will lead to electrical fires. Series fault arcs are the main cause of electrical fires. During their occurrence and continuous burning, the line current will decrease, which is difficult to be effectively sensed by traditional protection devices. Therefore, it is very necessary to detect series fault arcs to ensure the safety of power users.

[0003] In the prior art, the fault arc detection scheme usually uses the integral, variance or "zero-break" phenomenon of the time domain waveform of the current to determine whether a series fault arc occurs in the line. However, as the diversity of loads increases, signal confusion will become more serious. The actual series fault arc current waveform is confused with the normal operating current waveform of the power load. For example, the waveforms before and after the fault of power electronic switching devices are similar. Under normal operating conditions, the "zero-break" phenomenon will also appear at the zero crossing position of the waveform, resulting in confusion between the characteristics before and after the fault. Therefore, the fault arc detection method in the prior art is difficult to effectively distinguish the series fault arc signals under the current diverse load background, resulting in low detection accuracy and easy omission and misjudgment. Summary of the invention

[0004] The technical problem to be solved by the present invention is: in response to the technical problems existing in the prior art, the present invention provides a series fault arc detection method and device based on multi-dimensional feature fusion, which has a simple implementation method, low cost, high detection efficiency and accuracy, and low missed judgment and misjudgment rate.

[0005] In order to solve the above technical problems, the technical solution proposed by the present invention is: A series fault arc detection method based on multi-dimensional feature fusion, the steps comprising: Step S01, collecting line current signals and load end voltage data in real time; Step S02, extracting the wavelet energy entropy of the line current and the WRI feature used to characterize the periodic regularity of the waveform from the line current signal, calculating the arc gap voltage according to the load terminal voltage data, and fusing the extracted wavelet energy entropy of the line current, the WRI feature and the arc gap voltage to form a multi-dimensional discrimination feature of a series fault arc, wherein the WRI feature is calculated by extracting subsequences of the line current signal sequence that differ by one cycle and comprehensively calculating the differences between the subsequences; Step S03, input the extracted multi-dimensional discrimination features of the series fault arc into a pre-trained series fault arc detection model to identify whether there is a series fault arc at present. The series fault arc detection model is pre-trained based on a random forest model, using a series fault arc multi-dimensional discrimination feature data set composed of a combination of wavelet energy entropy, arc gap voltage and WRI features of the line current in a normal state and a series fault arc state under different load conditions.

[0006] Furthermore, in step S02, the multi-dimensional discrimination characteristics of the series fault arc include wavelet energy entropy of four layers d1-d4, the third and fifth harmonics of the arc gap voltage, and WRI characteristics. The arc gap voltage is calculated by subtracting the load terminal voltage from the power supply voltage.

[0007] Furthermore, the step of extracting the WRI feature includes: Acquire a sequence corresponding to the line current signal in a single time window as an original current sequence; Extracting a subsequence with a difference of one cycle from the original current sequence; Calculate a floating index based on the difference between the elements in each subsequence; The WRI feature is obtained by summing up all floating indices.

[0008] Further, from the original current sequence I Extract each subsequence according to the following formula:

[0009] Where n represents the number of subsequences to be extracted; No. j Floating index of subsequences The calculation expression is: ,…,

[0010] in, ~ Respectively represent No.1 in elements, ~ are the scaling factors used to amplify the differences at each point in the cycle.

[0011] Furthermore, when the number of elements in each subsequence is 3, the floating index The calculation expression is:

[0012] Each scaling factor w The calculation expression is:

[0013] Among them, A and B are w The two elements in the multiplication term.

[0014] Furthermore, in step S03, the training step of the series fault arc detection model includes: The wavelet energy entropy, arc gap voltage and WRI characteristics of the line current in the normal state and the series fault arc state under different load conditions are obtained, and data sampling processing is realized through a sliding time window, and a corresponding state code is assigned to each sample to construct the multi-dimensional discrimination feature data set of the series fault arc; The multi-dimensional discrimination feature data set of the series fault arc is divided into a training set and a test set in proportion, the feature vector of the training set is used as the model input, the state code is used as the model output, and the model is pre-trained by minimizing the model error; Selecting the optimal hyperparameter combination of the model according to the pre-training result, performing model training according to the optimal hyperparameter combination, and obtaining the series fault arc detection model after the training is completed; The series fault arc detection model is tested using test values.

[0015] Furthermore, in the process of training the series fault arc detection model based on the random forest model, it also includes the following steps: , Calculation time And the validation set accuracy The balance coefficient α is calculated as the number of weak classifiers changes, and the number of weak classifiers that makes the α value optimal is selected as the final number of weak classifiers to achieve the optimization of the number of weak classifiers. α The calculation method is:

[0016] in, , , Indicates weight, n_estimators value ranges from 1 to 1000, For each weak classifier number value, the model is cross-validated multiple times within the validation set to obtain the first statistical value of the output accuracy. The second statistical value of the output accuracy of the model obtained by cross-validating the output multiple times within the validation set for each weak classifier number value, For each weak classifier quantity value, the model is cross-validated multiple times within the validation set to obtain a statistical value of the output time consumed, wherein the first statistical value and the second statistical value are statistical values ​​of different types.

[0017] A series fault arc detection device based on multi-dimensional feature fusion, comprising: Data acquisition module, used to collect line current signals and load-end voltage data in real time; A multi-dimensional feature extraction module, used to extract the wavelet energy entropy of the line current from the line current signal and extract the WRI feature used to characterize the degree of regularity of the waveform period, calculate the arc gap voltage according to the load terminal voltage data, and fuse the extracted wavelet energy entropy of the line current, the WRI feature and the arc gap voltage to form a multi-dimensional discrimination feature of the series fault arc, wherein the WRI feature is calculated by extracting subsequences with a difference of one cycle in the line current signal sequence and comprehensively calculating the difference between each subsequence; A fault detection module is used to input the extracted multi-dimensional discrimination features of the series fault arc into a pre-trained series fault arc detection model to identify whether there is a series fault arc at present. The series fault arc detection model is pre-trained based on a random forest model using a series fault arc multi-dimensional discrimination feature data set composed of a combination of wavelet energy entropy, arc gap voltage and WRI features of the line current in a normal state and a series fault arc state under different load conditions.

[0018] An electronic device comprises a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to perform the above method.

[0019] A computer-readable storage medium storing a computer program, wherein the computer program implements the above method when executed by a processor.

[0020] Compared with the prior art, the advantages of the present invention are: the present invention forms a multi-dimensional discrimination feature of a series fault arc by comprehensively integrating arc gap voltage, wavelet energy entropy and waveform regularity characteristics, and forms a series fault detection model based on random forest model training by utilizing the multi-dimensional feature, which can comprehensively consider the characteristics that the high-frequency pulses of the arc are usually relatively concentrated and disordered while the pulses of the power electronic load are relatively uniform and regular, the characteristics that the arc gap voltage will specifically present a square wave-like characteristic when a series fault arc occurs under different load operating conditions, and the characteristics that the periodic regularity of the series fault arc signal is low, so that in the case of load diversity, the series fault arc can be distinguished quickly and accurately, and the problem that the detection accuracy of the traditional fault arc detection scheme is not high due to the similarity of the waveforms before and after the load fault of the power electronic switching device is solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a schematic diagram of the implementation flow of the series fault arc detection method based on multi-dimensional feature fusion in this embodiment.

[0022] Figure 2 It is a schematic diagram of a multi-dimensional discrimination feature data set of series fault arcs constructed in a specific application embodiment.

[0023] Figure 3 Schematic diagram of the principle of the random forest model used in this embodiment.

[0024] Figure 4 It is a flow chart of implementing series fault arc detection model training in this embodiment.

[0025] Figure 5 Schematic diagram of the principle of the sliding time window used in a specific application embodiment.

[0026] Figure 6 Schematic diagram of the principle of implementing hyperparameter selection in this embodiment. DETAILED DESCRIPTION

[0027] The present invention is further described below in conjunction with the accompanying drawings and specific preferred embodiments, but the protection scope of the present invention is not limited thereby.

[0028] As shown in the disclosure of the present invention, unless the context clearly indicates an exception, the words "a", "an", "a kind" and / or "the" do not specifically refer to the singular, but may also include the plural. The words "first", "second" and similar words used in the disclosure of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, the words "include" or "comprise" and the like mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects.

[0029] For ease of understanding, the relevant technical background of the present invention is first introduced by way of example.

[0030] For the detection of series fault arcs, the three main methods currently used are detection methods based on mathematical models, detection methods based on physical signals, and detection methods based on current / voltage waveform changes. Among them, the method based on mathematical models involves constructing dynamic relationship equations of arc voltage, current, power, energy and other parameters. In fact, through the establishment and application of the model, the fault arc identification process is guided and assisted. However, the arc detection method based on mathematical models will have problems such as the difference between the model and the actual situation, the verification of the applicability of the model, and the consideration of the diversity of arc current, which leads to the limitations of directly using traditional models for fault detection in practical applications. In addition, the fault arc detection method based on physical signals is mainly concentrated in the switch cabinets and distribution cabinets of substations and medium and low voltage distribution systems. The realization of detection needs to be highly dependent on precision sensors, such as sound, light, and thermal sensors, and needs to be designed and installed according to specific detection targets, which requires high costs. In addition, there will be problems such as weak signals may be covered by complex noise environments, limitations in sensor detection ranges, and uncertain positions, which contradict the randomness of fault arcs, resulting in low actual detection accuracy.

[0031] The detection method based on the change of current / voltage waveform mainly relies on the time domain and frequency domain characteristics of current / voltage, extracts the frequency domain and mathematical statistics characteristic values ​​of electrical data such as current and voltage before and after the fault, and combines artificial intelligence models to identify fault arcs. Common time domain features include integral and variance, and frequency domain features are obtained through Fourier transform methods. However, the above methods have problems such as complex parameter optimization and long calculation time. In addition, since the features used often lack diversity and representativeness, it is difficult to ensure that the pre-extracted fault arc features are representative, resulting in the low actual detection accuracy of such methods.

[0032] The waveforms before and after the load fault of power electronic switch devices are similar. In normal operation, the "zero rest" phenomenon will also appear at the zero crossing position of the waveform, resulting in confusion between the characteristics before and after the fault. Therefore, the diversity of loads will cause confusion between the normal state current signal and the fault arc current signal of power electronic switch devices, making it easy to misjudge or miss the detection process of series fault arcs. To solve the above problems, this embodiment first analyzes the characteristics of the series fault arc signal, and the series fault arc signal will have the following characteristics: (1) Under normal conditions, the waveforms of the load voltage and the power supply voltage are similar for different types of loads, and the difference between the power supply voltage and the load voltage is almost zero. When a series fault arc occurs, the arc gap voltage with square wave characteristics can be obtained by subtracting the load voltage from the power supply voltage. Therefore, the arc gap voltage can be used to distinguish the series fault arc signal. (2) The high-frequency pulses of the arc are usually relatively concentrated and disordered, while the pulses of the power electronic load are relatively uniform and regular. Therefore, based on this characteristic, the normal operation of the power electronic load and the arc state between different loads can be effectively distinguished, effectively reducing the misjudgment and missed judgment rate. The wavelet energy entropy is sensitive to signal mutations and can reflect the complexity of the signal sequence. It will change when the signal changes, and can characterize the changes in signal characteristics before and after the fault. Therefore, the wavelet energy entropy can also effectively distinguish the series fault arc signal.

[0033] (3) The current waveform of power electronic equipment under normal operation has a different degree of periodic regularity from the arc fault waveform. The arc waveform of a series fault is usually unstable and does not have periodic regularity. If the waveform has a high degree of regularity, it is more likely to be in normal operation. If the waveform has a low degree of regularity, it is more likely to be an arc fault. Therefore, the regularity of the waveform can also be used to identify the arc signal of a series fault.

[0034] Based on the above analysis, the present invention obtains line current data and load-end voltage data of normal state and series fault arc state under different load conditions, and then extracts arc gap voltage, wavelet energy entropy and WRI features to form a multidimensional, and forms a multidimensional discrimination feature of series fault arc by integrating arc gap voltage, wavelet energy entropy and waveform regularity features. The multidimensional feature is used to train a series fault detection model based on a random forest model, which can comprehensively consider the characteristics that the high-frequency pulses of the arc are usually relatively concentrated and disordered, while the pulses of the power electronic load are relatively uniform and regular, and the arc gap voltage will specifically present a square wave-like characteristic when a series fault arc occurs under different load operating conditions, and the characteristic that the regularity of the series fault arc signal period is low. In the case of load diversity, even if the waveforms before and after the load fault are similar, the series fault arc can still be quickly and accurately distinguished, thereby realizing efficient and accurate detection of the series fault arc, and solving the problem that the detection accuracy of the traditional fault arc detection scheme is low due to the similarity of the waveforms before and after the load fault of the power electronic switching device.

[0035] The present invention will be further described below in conjunction with specific embodiments.

[0036] like Figure 1 As shown, the series fault arc detection method based on multi-dimensional feature fusion in this embodiment includes the following steps: Step S01: collecting line current signals and load-end voltage data in real time.

[0037] Specifically, the user's real-time current data can be collected through the current sensor, the load end voltage can be obtained through the voltage transformer, and the arc gap voltage can be obtained by subtracting the power supply voltage from the load end voltage. That is, the arc gap voltage can be expressed as a voltage signal obtained by subtracting the power supply voltage from the load end voltage. The data is then normalized and further sampled using the sliding time window method. For example, the DS200UB-10V DS series sensor with high stability and high accuracy (ppm level) can be used and a 10kHz sampling rate can be set to collect line current signals, while also collecting load-end voltage and power supply voltage, and then the arc gap voltage is obtained by taking the difference between the power supply voltage and the load-end voltage.

[0038] Step S02, extracting the wavelet energy entropy of the line current and the WRI feature used to characterize the periodic regularity of the waveform from the line current signal, calculating the arc gap voltage according to the load-end voltage data, and fusing the extracted wavelet energy entropy of the line current, the WRI feature and the arc gap voltage to form a multi-dimensional discrimination feature of the series fault arc. The WRI feature is calculated by extracting subsequences of the line current signal sequence that differ by one cycle and comprehensively calculating the differences between the subsequences.

[0039] Considering that when a series fault arc occurs under different load operating conditions, the arc gap voltage presents square wave-like characteristics, and the odd harmonics in the square wave increase significantly, and the third and fifth harmonic contents increase significantly, which can effectively distinguish between normal and arc states; in view of the characteristics that the high-frequency pulses of the arc are relatively concentrated and disordered due to the diversity of loads, while the pulses of the power electronic load are relatively uniform and regular, the wavelet energy entropy of the four-layer detail signal of the line current signal d1-d4 can better characterize the above characteristics, so the line current d1-d4 wavelet energy entropy, the third and fifth harmonics of the arc gap voltage and the WRI characteristics are selected as the fault discrimination feature values ​​to constitute the multi-dimensional discrimination feature of the series fault arc, that is, the multi-dimensional discrimination feature of the series fault arc specifically includes the wavelet energy entropy of the four layers d1-d4, the third and fifth harmonics of the arc gap voltage and the WRI characteristics. By adopting the above-mentioned multi-dimensional discrimination feature of the series fault arc, this embodiment can explore the characteristics of the series fault arc signal as much as possible from a multi-dimensional perspective, and further improve the accuracy of the series fault arc detection.

[0040] In this embodiment, the step of extracting WRI features includes: Step S201. Acquire a sequence corresponding to a line current signal in a single time window as an original current sequence; Step S202. extracting a subsequence with a difference of one cycle from the original current sequence; Step S203. Calculate the floating index according to the difference between the elements in each subsequence; Step S204: Sum all floating indices to obtain the WRI feature.

[0041] Specifically, assuming that the original current sequence is I , each subsequence can be extracted according to the following formula: (1) Where n represents the number of subsequences that need to be extracted.

[0042] For the differences between all points that differ by several cycles, the greater the difference, the more unstable the sequence is. Therefore, the difference between the points of the cycle can characterize the stability and regularity of the sequence. This embodiment extracts all subsequences that differ by one cycle by using the time window sample period subsequence extraction method for the original line current signal sequence. Since the subsequences are obtained by extracting at the same interval of the original sequence, the elements in the subsequence can reflect the differences of the original sequence at the same phase, and avoid the interference of the inherent periodic fluctuations of the sequence in turn. For example, when the difference between a certain point and a point after several cycles is close to 0, it means that the sequence shows a stable periodic change with high regularity. Therefore, based on the difference between each element in the subsequence, the floating index that characterizes the floating state of the sequence fluctuation can be accurately calculated. The final WRI feature can be obtained by combining the calculated floating indexes of all subsequences, which can accurately reflect the degree of signal periodic regularity.

[0043] Specifically, the calculation of the WRI feature can be realized by constructing a floating index scaling factor w and amplifying the difference of the period points by weighted calculation. j Floating index of subsequences The calculation expression can be expressed as: ,…, (2) in, ~ Respectively represent No.1 in elements, ~ are scaling factors used to amplify the differences at each point in the cycle. If w is greater than 1, it can act as a weight to amplify the differences.

[0044] Each scaling factor w It can be calculated using the following expression: (3) Among them, A and B are w The two elements in the multiplication term, such as the calculation hour, A= 、B= .

[0045] From the above, we can see that the more regular the sequence is, the smaller the WRI value is. Conversely, the larger the WIR value is, the more irregular the sequence is, and the greater the probability is that it belongs to an arc fault.

[0046] The following example uses the case where the number of elements in each subsequence is 3 and processes the original data sequence using time window samples. is the original data sequence sample within a single time window. Extract I Subsequences with a difference of one period in , the first subsequence for: (4) in, Representative sequence I The first element in , and so on.

[0047] right I All subsequences that meet the above requirements are extracted to obtain: (5) Floating indicator The calculation expression is: (6) Finally, for all floating indicators F Sum and get the waveform regularity index WRI , that is: (7) Step S03, input the extracted multi-dimensional discrimination features of the series fault arc into a pre-trained series fault arc detection model to identify whether there is a series fault arc at present. The series fault arc detection model is pre-trained based on a random forest model, using a series fault arc multi-dimensional discrimination feature data set composed of a combination of wavelet energy entropy, arc gap voltage and WRI features of the line current in a normal state and a series fault arc state under different load conditions.

[0048] Specifically, in the model training stage, the fault arc under different load conditions is simulated through real experiments, and the current data of normal state and series fault arc state under different load conditions are collected. The line current d1-d4 wavelet energy entropy, the third and fifth harmonics of the arc gap voltage and the WRI features are extracted according to the same method as above, and a multi-dimensional discriminant feature data set of series fault arc is constructed. An integrated learning model suitable for series fault arc detection is built, and a series fault arc detection model is obtained by training the model with a large amount of data; in the real-time detection stage, the user's real-time current data is collected through the current sensor, the load end voltage is obtained by the voltage transformer, and the arc gap voltage is obtained by subtracting the power supply voltage from the load end voltage. The data is normalized and further sampled by the sliding time window method. Then, the wavelet energy entropy of the four-layer detail signal d1-d4 and the third and fifth harmonics of the arc gap voltage are extracted to construct a multi-dimensional feature vector, which is input into the trained series fault arc detection model based on the random forest algorithm, and whether a series fault arc occurs in the line is judged according to the model output result. For example, if the model outputs 1, it is judged that a series fault arc has occurred in the line, and if the model outputs 0, it is judged to be in a normal state.

[0049] A multi-dimensional discrimination feature data set for series fault arcs constructed in a specific application embodiment Figure 2 As shown (WRI features are not shown), it includes the wavelet energy entropy of the four-layer detail signals of the line current signal d1-d4 and the characteristic values ​​of the third and fifth harmonics of the arc gap voltage. In addition, it also includes the WRI features extracted based on the line current signal, which are all calculated from the data of nearly one sinusoidal cycle at the current sampling point.

[0050] In this embodiment, the training steps of the series fault arc detection model include: Step S301. Obtain the wavelet energy entropy, arc gap voltage and WRI characteristics of the line current in the normal state and the series fault arc state under different load conditions, and implement data sampling processing through a sliding time window, assign a corresponding state code to each sample, and construct a multi-dimensional discrimination feature data set for series fault arc; Step S302. Divide the multi-dimensional discrimination feature data set of the series fault arc into a training set and a test set in proportion, use the feature vector of the training set as the model input and the state code as the model output, and perform model pre-training by minimizing the model error; Step S303. Select the optimal hyperparameter combination of the model according to the pre-training result, perform model training according to the optimal hyperparameter combination, and obtain a series fault arc detection model after the training is completed; Step S304: Use the test value to test the series fault arc detection model.

[0051] The training principle of random forest model is as follows Figure 3As shown in the figure, the model uses Booststrap sampling to uniformly sample from the training set with replacement, and the random forest uses the bootstrap resampling technique to repeatedly randomly select n samples (n is generally the same as N) from the original training sample set N with replacement to generate a new training sample set, and then generates k classification trees based on the bootstrap sample set to form a random forest. The model adopts the idea of ​​"brainstorming" and combines multiple decision trees for ensemble learning. Each decision tree outputs a classification voting result. By integrating the classification voting results of all decision trees and taking the classification result with the largest number of votes as the final output result, the classification prediction result is closer to the actual value.

[0052] In a specific application embodiment, the overall training process of the series fault arc detection model is as follows: Figure 4 As shown in the figure, normal and fault arc current data under different working conditions are obtained through real experiments, and data sampling is further realized through sliding time windows. Each sample is assigned a corresponding state code to complete the construction of the arc discrimination feature data set; the data set is divided into training set and test set in a ratio of 4:1, and the feature vector of the training set is used as the model input, and the state code is used as the model output. The GridSearch method is used to minimize the model error for pre-training, and the optimal hyperparameter combination of the model is selected. Then, the model is formally trained to complete the construction of the series fault arc detection model based on the random forest algorithm, and the model effectiveness is tested through the test set.

[0053] The sliding time window method used in the specific application embodiment is as follows: Figure 5 As shown, at a sampling rate of 10kHz, with a window length of one cycle (200 sampling points), sliding sampling is performed in sequence to obtain data. Each window of data is used to calculate the characteristic value of the current sampling point and corresponds to the state at the current sampling moment.

[0054] This embodiment further selects the optimal hyperparameter combination of the model by utilizing a hyperparameter optimization algorithm to achieve the optimal detection level, thereby realizing accurate detection of series fault arcs and improving the accuracy of series fault arc detection.

[0055] The hyperparameter selection process in a specific application embodiment is as follows: Figure 6As shown in the figure, the GridSearch hyperparameter optimization method is used to obtain models under different hyperparameter combinations, and the error of each model is evaluated respectively, and finally the model corresponding to the optimal hyperparameter combination is found. Specifically, based on the GridSearch method, a hyperparameter sample vector is formed by cross-combining in the hyperparameter search domain, and the candidate parameters in the domain are sampled in turn, and a temporary verification model is constructed. The hyperparameter sample vector is evaluated for accuracy based on the verification set data. The evaluation results are matched with the corresponding hyperparameter sample vectors in turn and sorted one by one, and finally the locally optimal hyperparameter sample combination in the hyperparameter search domain is obtained, which effectively avoids the problem of falling into local extreme values ​​during radial neural network training and ensures the model operation performance.

[0056] In the process of training the series fault arc detection model based on the random forest model, this embodiment also includes the following steps: , Calculation time And the validation set accuracy The balance coefficient α is calculated as the number of weak classifiers n_estimators changes, and the number of weak classifiers that makes the α value optimal is selected as the final number of weak classifiers to achieve the optimization of the number of weak classifiers and the balance coefficient α The calculation method is: (8) in, , , represents the weight, For each weak classifier number value, the model is cross-validated multiple times within the validation set to obtain the first statistical value of the output accuracy. The second statistical value of the output accuracy of the model obtained by cross-validating the output multiple times within the validation set for each weak classifier number value, For each weak classifier quantity value, the model cross-validates the output multiple times within the validation set to obtain a statistical value of the output time consumption, and the first statistical value and the second statistical value are different types of statistical values, such as variance, mean, etc.

[0057] The detection of series arcs requires high timeliness and reliability. In order to match the timeliness and reliability of series arc detection, the model needs to ensure accuracy while ensuring that the calculation time is short enough and the model output results are relatively stable. This embodiment uses the above-mentioned weak classifier number optimization method based on the balance coefficient α combined with the validation set random multiple cross-validation model performance test to accurately select the optimal number of weak classifiers. It can effectively balance the accuracy, stability and calculation time in the process of selecting the number of weak classifiers of the random forest model, so that the final model can take into account the model accuracy, stability and calculation time, and avoid the situation where the number of weak classifiers is too high and the model calculation time is too long when only considering stability and accuracy, thereby effectively ensuring the timeliness and reliability of the series fault arc detection model.

[0058] In a specific application embodiment, , , The values ​​can be -1.3, 1.1, and -1.1 respectively; the number of weak classifiers n_estimators is from 1 to 1000, and the model is cross-validated 100 times within the validation set at each n_estimators value. is the variance of the 100 output accuracies, which is used to reflect the stability of the model under the current n_estimators value. is the average time consumed for these 100 outputs, The average value of the 100 output accuracies is finally selected as the value of n_estimators that maximizes the α value as the number of weak classifiers in the model hyperparameters. Through the above method, the model performance test mode based on random multiple cross-validation of the validation set can be used to achieve balanced optimization of the number of weak classifiers, which is helpful for optimizing the overall hyperparameters of the random forest model.

[0059] The present invention selects features that have a strong correlation with the occurrence of series fault arcs based on whether the pulse signal is disordered and whether the arc gap voltage has square wave-like characteristics when a series fault arc occurs under different load operating states, thereby effectively improving the detection accuracy and reducing the risks of misjudgment and missed judgment. It can effectively distinguish between different load fault arc conditions and normal load conditions of power electronic switch devices, and the extracted features are universal. It can be achieved by adopting a sampling rate of, for example, 10kHz, avoiding the use of high-frequency sampling devices, and effectively reducing costs. Combined with the model training method of the random forest model, it can also reduce model training and calculation time while improving the fault detection accuracy.

[0060] The series fault arc detection device based on multi-dimensional feature fusion in this embodiment includes: Data acquisition module, used to collect line current signals and load-end voltage data in real time; The multi-dimensional feature extraction module is used to extract the wavelet energy entropy of the line current from the line current signal and extract the WRI feature used to characterize the regularity of the waveform period, calculate the arc gap voltage according to the load terminal voltage data, and fuse the extracted wavelet energy entropy of the line current, the WRI feature and the arc gap voltage to form a multi-dimensional discrimination feature of the series fault arc. The WRI feature is calculated by extracting subsequences with a difference of one cycle in the line current signal sequence and comprehensively calculating the difference between each subsequence; The fault detection module is used to input the extracted multi-dimensional discrimination features of the series fault arc into a pre-trained series fault arc detection model to identify whether there is a series fault arc at present. The series fault arc detection model is pre-trained based on a random forest model using a series fault arc multi-dimensional discrimination feature data set composed of wavelet energy entropy, arc gap voltage and WRI features of line current in normal state and series fault arc state under different load conditions. The series fault arc detection device based on multi-dimensional feature fusion in this embodiment corresponds one to one with the above-mentioned series fault arc detection method based on multi-dimensional feature fusion, and will not be described one by one here.

[0061] This embodiment further provides an electronic device, including a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to perform the above method.

[0062] It is understandable that the above method of this embodiment can be executed by a single device, such as a computer or server, etc., and can also be applied to a distributed scenario and completed by multiple devices in cooperation with each other. In the case of a distributed scenario, one of the multiple devices can only execute one or more steps in the above method of this embodiment, and multiple devices interact to complete the above method. The processor can be implemented in the form of a general-purpose CPU, a microprocessor, an application-specific integrated circuit, or one or more integrated circuits, etc., for executing related programs to implement the above method of this embodiment. The memory can be implemented in the form of a read-only memory ROM, a random access memory RAM, a static storage device, and a dynamic storage device. The memory can store an operating system and other applications. When the above method of this embodiment is implemented by software or firmware, the relevant program code is stored in the memory and called and executed by the processor.

[0063] This embodiment further provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the above method is implemented.

[0064] Those skilled in the art should understand that the above-mentioned embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application can take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the process Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0065] The above is only a preferred embodiment of the present invention, and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment, it is not intended to limit the present invention. Therefore, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for detecting series fault arc based on multi-dimensional feature fusion, characterized in that the steps include: Step S01, collecting line current signals and load end voltage data in real time; Step S02, extracting the wavelet energy entropy of the line current and the WRI feature used to characterize the periodic regularity of the waveform from the line current signal, calculating the arc gap voltage according to the load terminal voltage data, and fusing the extracted wavelet energy entropy of the line current, the WRI feature and the arc gap voltage to form a multi-dimensional discrimination feature of a series fault arc, wherein the WRI feature is calculated by extracting subsequences of the line current signal sequence that differ by one cycle and comprehensively calculating the differences between the subsequences; Step S03, input the extracted multi-dimensional discrimination features of the series fault arc into a pre-trained series fault arc detection model to identify whether there is a series fault arc at present. The series fault arc detection model is pre-trained based on a random forest model, using a series fault arc multi-dimensional discrimination feature data set composed of a combination of wavelet energy entropy, arc gap voltage and WRI features of the line current in a normal state and a series fault arc state under different load conditions.

2. The method for detecting series fault arc based on multi-dimensional feature fusion according to claim 1, characterized in that: In step S02, the multi-dimensional discrimination features of the series fault arc include wavelet energy entropy of four layers d1-d4, the third and fifth harmonics of the arc gap voltage, and WRI features. The arc gap voltage is calculated by subtracting the load terminal voltage from the power supply voltage.

3. The method for detecting series fault arc based on multi-dimensional feature fusion according to claim 1, characterized in that: The steps of extracting the WRI features include: Acquire a sequence corresponding to the line current signal in a single time window as an original current sequence; Extracting a subsequence with a difference of one cycle from the original current sequence; Calculate a floating index based on the difference between the elements in each subsequence; The WRI feature is obtained by summing up all floating indices.

4. The method for detecting series fault arc based on multi-dimensional feature fusion according to claim 3, characterized in that: From the original current sequence I Extract each subsequence according to the following formula: Where n represents the number of subsequences to be extracted; No. j Floating index of subsequences The calculation expression is: ,…, in, ~ Respectively represent No.1~ elements, ~ are the scaling factors used to amplify the differences at each point in the cycle.

5. The method for detecting series fault arc based on multi-dimensional feature fusion according to claim 4, characterized in that: When the number of elements in each subsequence is 3, the floating index The calculation expression is: Each scaling factor w The calculation expression is: Among them, A and B are w The two elements in the multiplication term.

6. The method for detecting series fault arc based on multi-dimensional feature fusion according to any one of claims 1 to 5, characterized in that: In step S03, the training steps of the series fault arc detection model include: The wavelet energy entropy, arc gap voltage and WRI characteristics of the line current in the normal state and the series fault arc state under different load conditions are obtained, and data sampling processing is realized through a sliding time window, and a corresponding state code is assigned to each sample to construct the multi-dimensional discrimination feature data set of the series fault arc; The multi-dimensional discrimination feature data set of the series fault arc is divided into a training set and a test set in proportion, the feature vector of the training set is used as the model input, the state code is used as the model output, and the model is pre-trained by minimizing the model error; Selecting the optimal hyperparameter combination of the model according to the pre-training result, performing model training according to the optimal hyperparameter combination, and obtaining the series fault arc detection model after the training is completed; The series fault arc detection model is tested using test values.

7. The method for detecting series fault arc based on multi-dimensional feature fusion according to any one of claims 1 to 5, characterized in that: In the process of training the series fault arc detection model based on the random forest model, it also includes the stability of the model output. , Calculation time And the validation set accuracy The balance coefficient α is calculated as the number of weak classifiers changes, and the number of weak classifiers that makes the α value optimal is selected as the final number of weak classifiers to achieve the optimization of the number of weak classifiers. α The calculation method is: in, , , Indicates weight, n_estimators value ranges from 1 to 1000, For each weak classifier number value, the model is cross-validated multiple times within the validation set to obtain the first statistical value of the output accuracy. The second statistical value of the output accuracy of the model obtained by cross-validating the output multiple times within the validation set for each weak classifier number value, For each weak classifier quantity value, the model is cross-validated multiple times within the validation set to obtain a statistical value of the output time consumed, wherein the first statistical value and the second statistical value are statistical values ​​of different types.

8. A series fault arc detection device based on multi-dimensional feature fusion, characterized in that: include: Data acquisition module, used to collect line current signals and load-end voltage data in real time; A multi-dimensional feature extraction module, used to extract the wavelet energy entropy of the line current from the line current signal and extract the WRI feature used to characterize the degree of regularity of the waveform period, calculate the arc gap voltage according to the load terminal voltage data, and fuse the extracted wavelet energy entropy of the line current, the WRI feature and the arc gap voltage to form a multi-dimensional discrimination feature of the series fault arc, wherein the WRI feature is calculated by extracting subsequences with a difference of one cycle in the line current signal sequence and comprehensively calculating the difference between each subsequence; A fault detection module is used to input the extracted multi-dimensional discrimination features of the series fault arc into a pre-trained series fault arc detection model to identify whether there is a series fault arc at present. The series fault arc detection model is pre-trained based on a random forest model using a series fault arc multi-dimensional discrimination feature data set composed of a combination of wavelet energy entropy, arc gap voltage and WRI features of the line current in a normal state and a series fault arc state under different load conditions.

9. An electronic device comprising a processor and a memory, wherein the memory is used to store a computer program, wherein: The processor is configured to execute the computer program to perform the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images