Multi-feature abnormal arc detection method, ARM control module and device

Abstract

The invention discloses a multi-feature abnormal arc detection method, an ARM control module and a device, and relates to the field of electrical safety monitoring, and the method comprises the steps: obtaining current information in a power line; performing data conversion calibration processing on the current information to generate target data; caching the target data of the plurality of periods and extracting a feature combination corresponding to a pre-trained abnormal arc detection model from the target data of each period, wherein the feature combination comprises any one or combination of a time domain feature, a frequency domain feature, a time frequency feature and a modal feature; inputting the feature combination of each period into a pre-trained abnormal arc detection model for detection so as to output an arc detection result of each period; and judging an abnormal arc according to the arc detection result of each period. According to the invention, the accuracy of abnormal arc detection can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of electrical safety monitoring, and in particular to a multi-feature abnormal arc detection method, an ARM control module and a device. Background Art

[0002] At present, the core technology of technical solutions and products for detecting fault arcs is mainly based on identification based on electrical parameters (such as current signals and voltage signals), and the detection model is trained by extracting corresponding electrical parameter features to achieve the purpose of abnormal arc detection.

[0003] However, the number of electrical parameter features used in the prior art is usually one or two. For example, Chinese patent CN114034960A discloses a fault arc detection method based on weak arc features, which only detects fault arcs through the time domain characteristics of the arc current waveform.

[0004] However, the small number of electrical parameter features will cause some data information to be lost, which can easily lead to low detection accuracy. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a multi-feature abnormal arc detection method, ARM control module and device, which can effectively improve the accuracy of abnormal arc detection.

[0006] In order to solve the above technical problems, the present invention provides a multi-feature abnormal arc detection method, including: obtaining current information in a power line; performing data conversion and calibration processing on the current information to generate target data; caching target data of multiple cycles and extracting feature combinations corresponding to a pre-trained abnormal arc detection model from the target data of each cycle, respectively, the feature combination including any one or a combination of time domain features, frequency domain features, time-frequency features and modal features; inputting the feature combination of each cycle into a pre-trained abnormal arc detection model for detection to output arc detection results of each cycle; judging the abnormal arc according to the arc detection results of each cycle.

[0007] As an improvement of the above scheme, the time domain characteristics include mean, variance, standard deviation, root mean square, peak-to-peak value, skewness, kurtosis, waveform factor, peak factor, pulse factor, margin factor, kurtosis factor and energy index.

[0008] As an improvement of the above solution, the step of calculating the frequency domain feature includes:

[0009] Perform Fourier transform on the current information to generate a Fourier transform result, wherein the Fourier transform result is D fft =|FFT(i)|;

[0010] The Fourier transform result is divided into m parts to form a target matrix, and the target matrix is ​​D fft [m] = D fft / m;

[0011] According to the formula Calculate frequency domain features;

[0012] Where, FFT(*) is Fourier transform, D fft is the Fourier transform result, D fft [m] is the target matrix, fft[m] f It is the frequency domain feature.

[0013] As an improvement of the above scheme, the step of calculating the time-frequency characteristics includes: performing wavelet transform on the current information to generate a wavelet transform result, wherein the wavelet transform result is Dwad=WT(i,5); according to formula k max =min(len(Dwad[j])), k=0,1,2,…,5, calculate the shortest subband length in the wavelet transform result; according to the formula power(k)=|Dwad[j][k]|, k=0,1,2,…,k max , calculate the time-frequency characteristics; where WT(i,5) is the wavelet transform with input data i and layer number 5, Dwad is the wavelet transform result, k max is the shortest subband length in the wavelet transform result, and power(k) is the time-frequency feature.

[0014] As an improvement of the above solution, the step of extracting the modal features includes:

[0015] Perform empirical mode decomposition on the current information to generate a decomposition result, wherein the decomposition result is IMF i =EMD(s i );

[0016] According to the formula Calculate modal characteristics;

[0017] Among them, EMD(*) is the EMD algorithm, IMF i is the decomposition result, imf_feature is the modal feature, and F is the number of decomposition levels.

[0018] As an improvement of the above-mentioned scheme, the step of judging an abnormal arc based on the arc detection results of each cycle includes: counting the arc detection results of each cycle; when the number of cycles in which the arc detection result is "arc" is greater than a preset number of times, it is judged that "abnormal arc occurs", otherwise, it is judged that "no abnormal arc occurs"; or when the ratio of the number of cycles in which the arc detection result is "arc" to the total number of cycles is greater than a preset ratio, it is judged that "abnormal arc occurs", otherwise, it is judged that "no abnormal arc occurs".

[0019] As an improvement of the above scheme, the step of constructing the feature combination includes: providing a feature set, the feature elements in the feature set include time domain features, frequency domain features, time-frequency features and modal features; sorting the feature elements by feature importance; according to the feature importance sorting result, finding the combination of feature elements corresponding to the classification model with the highest accuracy through an iterative optimization algorithm; using the classification model with the highest accuracy as the trained abnormal arc detection model, and using the combination of feature elements corresponding to the classification model with the highest accuracy as the feature combination corresponding to the trained abnormal arc detection model.

[0020] As an improvement of the above solution, the classification model is a KNN classification model.

[0021] Correspondingly, the present invention also provides an ARM control module, including a memory and a processor, wherein the memory stores a computer program, wherein the processor implements the steps of the above-mentioned multi-feature abnormal arc detection method when executing the computer program.

[0022] Correspondingly, the present invention also provides a multi-feature abnormal arc detection device, characterized in that it includes a current sampling module, the above-mentioned ARM control module, a display screen and a power supply module, the ARM control module is respectively connected to the current sampling module and the display screen; the current sampling module is used to collect current information in the power line; the display screen is used to display the detection result of the ARM control module; the power supply module is respectively connected to the current sampling module, the ARM control module and the display screen, and is used to supply power to the current sampling module, the ARM control module and the display screen.

[0023] The implementation of the present invention has the following beneficial effects:

[0024] The multi-feature abnormal arc detection method of the present invention is based on multiple feature elements such as time domain features, frequency domain features, time-frequency features and modal features, and sorts the feature elements by importance, so as to select a feature combination training classification model with the highest detection accuracy and a relatively small number of features, so as to achieve the purpose of improving the accuracy of abnormal arc detection;

[0025] At the same time, in the multi-feature abnormal arc detection method of the present invention, the feature combination and the abnormal arc detection model correspond one to one, and the specific combination method is determined during the training of the abnormal arc detection model, which is highly targeted. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a flow chart of an embodiment of a method for detecting abnormal arc with multiple features of the present invention;

[0027] Figure 2 is a flow chart of another embodiment of the abnormal arc detection method with multiple features of the present invention;

[0028] Figure 3 It is a schematic diagram of the structure of an embodiment of the abnormal arc detection device with multiple features of the present invention. DETAILED DESCRIPTION

[0029] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings.

[0030] See also Figure 1 , Figure 1 The flowchart of an embodiment of the abnormal arc detection method of the present invention with multiple features is shown, which includes:

[0031] S101, obtaining current information in the power line;

[0032] S102, performing data conversion and calibration processing on the current information to generate target data;

[0033] After data conversion and calibration processing is performed on the current information, accurate floating-point current data, namely, target data, can be obtained.

[0034] S103, caching target data of multiple cycles and extracting feature combinations corresponding to the pre-trained abnormal arc detection model from the target data of each cycle;

[0035] Generally, the number of cache cycles can be set according to the actual situation, and preferably 15 to 50 cycles of target data can be cached. When the number of cache cycles is less than 15, the accuracy of the detection result will be affected, and when the number of cache cycles is greater than 50, too much memory will be occupied.

[0036] It should be noted that the abnormal arc detection model corresponds to the feature combination one by one; wherein the feature combination includes any one or combination of time domain features, frequency domain features, time-frequency features and modal features, that is, the feature combination can be a single frequency domain feature, a combination of frequency domain features and time-frequency features, or a combination of time-frequency features and modal features... The specific combination method is determined during the training of the abnormal arc detection model and is highly targeted.

[0037] The time domain features, frequency domain features, time-frequency features and modal features are described below:

[0038] 1. Time Domain Characteristics

[0039] Time domain characteristics include mean, variance, standard deviation, root mean square, peak-to-peak value, skewness, kurtosis, shape factor, peak factor, impulse factor, margin factor, kurtosis factor and energy index. Specifically:

[0040] The mean is expressed as:

[0041]

[0042] The variance is expressed as:

[0043]

[0044] The standard deviation is expressed as:

[0045]

[0046] The root mean square is expressed as:

[0047]

[0048] The peak-to-peak value is expressed as:

[0049] peak2peak=s max -s min

[0050] Skewness is expressed as:

[0051]

[0052] Kurtosis is expressed as:

[0053]

[0054] The shape factor is expressed as:

[0055]

[0056] The crest factor is expressed as:

[0057]

[0058] The impulse factor is expressed as:

[0059]

[0060] The margin factor is expressed as:

[0061]

[0062] The kurtosis factor is expressed as:

[0063]

[0064] The energy index is expressed as:

[0065]

[0066] Among them, S i is the time series data (i.e., current information), and N is the time corresponding to the current information.

[0067] 2. Frequency Domain Characteristics

[0068] Accordingly, the calculation steps of frequency domain features include:

[0069] (1) According to the following formula, the current information is Fourier transformed to generate the Fourier transform result D fft :

[0070] D fft =|FFT(i)|

[0071] Among them, FFT(*) is Fourier transform.

[0072] (2) According to the following formula, the Fourier transform result is divided into m parts to form the target matrix D fft [m]:

[0073] D fft [m] = D fft / m

[0074] (3) Calculate the frequency domain feature fft[m] according to the following formula: f :

[0075]

[0076] 3. Time-frequency characteristics

[0077] Accordingly, the calculation steps of time-frequency features include:

[0078] (1) According to the following formula, the current information is subjected to wavelet transform to generate a wavelet transform result Dwad:

[0079] Dwad=WT(i,5)

[0080] in,

[0081] WT(*) is the wavelet transform, WT(i,5) is the wavelet transform with input data i and 5 layers;

[0082] Dwad is the result of wavelet transform, including 1 layer of approximate parameters and 5 layers of detail parameters, i.e. the content of j.

[0083] (2) According to the following formula, calculate the shortest subband length k in the wavelet transform result max :

[0084] k max =min(len(Dwad[j])),k=0,1,2,…,5

[0085] According to the formula, calculate the time-frequency characteristics;

[0086] (3) Calculate the time-frequency feature power(k) according to the following formula:

[0087] power(k)=|Dwad[j][k]|,k=0,1,2,…,k max

[0088] Among them, power(k) is the absolute value matrix of approximate parameters and detail parameters.

[0089] For example, if the input current data is 1024 points, after wavelet transform, the detailed parameters of 512, 256, 128, 64, and 32 points and the approximate parameters of 32 points can be obtained. By calculating the sum of squares for each parameter, 6 time-frequency features can be obtained.

[0090] 4. Modal Characteristics

[0091] Accordingly, the steps of extracting modal features include:

[0092] (1) According to the following formula, the current information is subjected to empirical mode decomposition to generate the decomposition result IMF i :

[0093] IMF i =EMD(s i )

[0094] Among them, EMD(*) is the EMD algorithm;

[0095] (2) Calculate the modal feature imf_feature according to the following formula:

[0096]

[0097] Wherein, F is the number of decomposed layers, and in this embodiment, F=8.

[0098] like Figure 2 As shown, in the present invention, the construction of the feature combination is carried out simultaneously with the training of the abnormal arc detection model. Accordingly, the steps of constructing the feature combination include:

[0099] (1) Propose a feature set;

[0100] The feature elements in the feature set include time domain features, frequency domain features, time-frequency features and modal features;

[0101] (2) Sort the feature elements by their importance;

[0102] Preferably, the Pearson algorithm may be used to rank the feature elements according to their feature importance.

[0103] (3) Based on the feature importance ranking results, an iterative optimization algorithm is used to find the combination of feature elements corresponding to the classification model with the highest accuracy;

[0104] (4) The classification model with the highest accuracy is used as the trained abnormal arc detection model, and the combination of feature elements corresponding to the classification model with the highest accuracy is used as the feature combination corresponding to the trained abnormal arc detection model.

[0105] Preferably, the classification model may be a KNN (nearest neighbor classification) classification model.

[0106] That is to say, after the present invention performs data conversion and calibration processing on the collected current information in the power lines, characteristic elements (mean, variance, standard deviation, root mean square, peak-to-peak value, skewness, kurtosis, waveform factor, peak factor, pulse factor, margin factor, kurtosis factor, energy index, frequency domain characteristics, time-frequency characteristics and modal characteristics) are extracted from the current information, and then the characteristic elements are sorted by feature importance, and the combination of characteristic elements is adjusted in turn according to the feature importance sorting structure to perform model training, and the optimal feature combination is found (the optimization goal is to maximize the accuracy of the classification model), and then the classification model trained by the feature combination is saved for detection.

[0107] S104, inputting the feature combination of each cycle into a pre-trained abnormal arc detection model for detection, so as to output the arc detection result of each cycle;

[0108] Correspondingly, the present invention performs data conversion calibration on the acquired current data in the same way as the training part, caches the current data of multiple cycles, extracts the feature combination obtained with the training part, and inputs the abnormal arc detection model to obtain the arc detection results of each cycle.

[0109] S105, judging abnormal arc according to the arc detection result of each cycle.

[0110] Accordingly, the steps of determining abnormal arcs according to the arc detection results of each cycle include:

[0111] (1) Count the arc detection results of each cycle;

[0112] (2) When the number of cycles of the arc detection result being "arc" is greater than the preset number, it is judged as "abnormal arc occurs", otherwise, it is judged as "abnormal arc does not occur";

[0113] For example, when there are 20 cycles, the preset number may be set to 10.

[0114] (3) When the ratio of the number of cycles with the arc detection result of "arc" to the total number of cycles is greater than a preset ratio, it is judged that "abnormal arc occurs", otherwise, it is judged that "abnormal arc does not occur".

[0115] For example, when there are 20 cycles, the preset ratio may be set to 1 / 2.

[0116] It should be noted that the use of statistical knowledge in the detection stage can avoid misdetecting a non-abnormal arc (an arc that may be generated when the appliance is normally started) as an abnormal arc. The present invention counts the number of cycles in which the arc occurs, and when the number of cycles is greater than a certain number / ratio, it is considered that an abnormal arc has occurred.

[0117] Furthermore, the final abnormal arc judgment result can be displayed on a display screen and processed by management personnel in a timely manner.

[0118] From the above, it can be seen that the present invention is based on multiple characteristic elements such as mean, variance, standard deviation, root mean square, peak-to-peak value, skewness, kurtosis, waveform factor, peak factor, pulse factor, margin factor, kurtosis factor, energy index, frequency domain characteristics, time-frequency characteristics and modal characteristics, and sorts the characteristic elements by importance, so as to screen out the feature combination with the highest detection accuracy and a relatively small number of features to train the classification model, so as to achieve the purpose of improving the accuracy of abnormal arc detection.

[0119] See also Figure 3 , Figure 3 The specific structure of the abnormal arc detection device 100 with multiple features of the present invention is shown, which includes a current sampling module 1, an ARM control module 2, a display screen 3 and a power supply module 4, wherein the ARM control module 2 is connected to the current sampling module 1 and the display screen 3 respectively;

[0120] The current sampling module 1 is used to collect current information in the power line;

[0121] Display screen 3 is used to display the detection results of the ARM control module;

[0122] An ARM control module 2 includes a memory and a processor, wherein the memory stores a computer program, wherein the processor implements the steps of the above-mentioned multi-feature abnormal arc detection method when executing the computer program;

[0123] The power supply module 4 is connected to the current sampling module 1 , the ARM control module 2 and the display screen 3 respectively, and is used to supply power to the current sampling module 1 , the ARM control module 2 and the display screen 3 .

[0124] It should be noted that the multi-feature abnormal arc detection device 100 can be installed in a distribution box. The current sampling module 1 collects the current information in the power line in real time and sends the current information to the ARM control module 2; the ARM control module 2 performs data conversion and calibration processing on the current information to generate target data. When the ARM control module 2 caches enough target data for a period of time, it extracts feature combinations from the target data of each period, and inputs the feature combinations into the pre-trained abnormal arc detection model for detection, and outputs the arc detection results of each period. Then, according to the arc detection results of each period, the abnormal arc can be judged; finally, the ARM control module 2 sends the detection results to the display screen 3 for display, so that it can be processed by the management personnel in time.

[0125] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A multi-feature abnormal arc detection method, characterized in that: include: Get the current information in the power line; Performing data conversion and calibration processing on the current information to generate target data; Cache target data of multiple cycles and extract feature combinations corresponding to a pre-trained abnormal arc detection model from the target data of each cycle, wherein the feature combinations include any one or a combination of time domain features, frequency domain features, time-frequency features, and modal features; Inputting the feature combination of each cycle into a pre-trained abnormal arc detection model for detection to output the arc detection result of each cycle; The abnormal arc is judged based on the arc detection results of each cycle.

2. The multi-feature abnormal arc detection method according to claim 1, characterized in that: The time domain characteristics include mean, variance, standard deviation, root mean square, peak-to-peak value, skewness, kurtosis, waveform factor, peak factor, pulse factor, margin factor, kurtosis factor and energy index.

3. The multi-feature abnormal arc detection method according to claim 1, characterized in that: The calculation steps of the frequency domain features include: Perform Fourier transform on the current information to generate a Fourier transform result, wherein the Fourier transform result is D fft =|FFT(i)|; The Fourier transform result is divided into m parts to form a target matrix, and the target matrix is ​​D fft [m] = D fft / m; According to the formula Calculate frequency domain features; Where, FFT(*) is Fourier transform, D fft is the Fourier transform result, D fft [m] is the target matrix, fft[m] f is the frequency domain feature.

4. The multi-feature abnormal arc detection method according to claim 1, characterized in that: The step of calculating the time-frequency feature includes: Performing wavelet transform on the current information to generate a wavelet transform result, wherein the wavelet transform result is Dwad=WT(i,5); According to the formula k max =min(len(Dwad[j])), k=0,1,2,…,5, calculate the shortest subband length in the wavelet transform result; According to the formula power(k)=|Dwad[j][k]|, k=0,1,2,…,k max , calculate time-frequency features; Among them, WT(i,5) is the wavelet transform with input data i and layer number 5, Dwad is the wavelet transform result, k max is the shortest subband length in the wavelet transform result, and power(k) is the time-frequency feature.

5. The multi-feature abnormal arc detection method according to claim 1, characterized in that: The step of extracting the modal features comprises: Perform empirical mode decomposition on the current information to generate a decomposition result, wherein the decomposition result is IMF i =EMD(s i ); According to the formula Calculate modal characteristics; Among them, EMD(*) is the EMD algorithm, IMF i is the decomposition result, imf_feature is the modal feature, and F is the number of decomposition levels.

6. The multi-feature abnormal arc detection method according to claim 1, characterized in that: The step of determining abnormal arcs according to the arc detection results of each cycle includes: Count the arc detection results of each cycle; When the number of cycles of the arc detection result being "arc" is greater than the preset number, it is judged as "abnormal arc occurs", otherwise, it is judged as "abnormal arc does not occur"; or When the ratio of the number of cycles in which the arc detection result is "arc" to the total number of cycles is greater than a preset ratio, it is judged as "abnormal arc occurs", otherwise, it is judged as "no abnormal arc occurs".

7. The multi-feature abnormal arc detection method according to claim 1, characterized in that: The steps of constructing the feature combination include: Providing a feature set, wherein the feature elements in the feature set include time domain features, frequency domain features, time-frequency features and modal features; Sorting the characteristic elements according to their importance; According to the result of feature importance sorting, the combination of feature elements corresponding to the classification model with the highest accuracy is found through an iterative optimization algorithm; The classification model with the highest accuracy is used as the trained abnormal arc detection model, and the combination of feature elements corresponding to the classification model with the highest accuracy is used as the feature combination corresponding to the trained abnormal arc detection model.

8. The multi-feature abnormal arc detection method according to claim 7, characterized in that: The classification model is a KNN classification model.

9. An ARM control module, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A multi-feature abnormal arc detection device, characterized in that: It comprises a current sampling module, the ARM control module according to claim 9, a display screen and a power supply module, wherein the ARM control module is connected to the current sampling module and the display screen respectively; The current sampling module is used to collect current information in the power line; The display screen is used to display the detection result of the ARM control module; The power supply module is connected to the current sampling module, the ARM control module and the display screen respectively, and is used to supply power to the current sampling module, the ARM control module and the display screen.

Images