Multi-modal fusion fault arc detection method, computer equipment and multi-modal fusion fault arc detection system

Abstract

The invention discloses a multi-modal fusion fault arc detection method, computer equipment and system, and relates to the field of electrical safety monitoring, and the method comprises the steps: obtaining multiple groups of modal information under a target working condition; preprocessing the multiple groups of modal information to extract multiple groups of modal feature information; generating multiple groups of modal fault identification results according to the multiple groups of modal feature information, wherein the modal information, the modal feature information and the modal fault identification results are in one-to-one correspondence; and fusing the multiple groups of modal fault recognition results into a target fault recognition result to determine whether a fault arc exists in the target working condition. According to the invention, the false alarm rate of the arc identification model can be effectively reduced, and the reliability of the arc fault detection system is improved.

Description

Technical Field

[0001] The present invention relates to the field of electrical safety monitoring, and in particular to a multi-modal fusion fault arc detection method, computer equipment and system. Background Art

[0002] Most of the existing arc fault detection methods are based on the waveform characteristics, frequency, amplitude or frequency domain characteristics of the current signal as the basis for identifying the arc fault; however, the above arc detection methods are based on the extraction of features from a single physical signal as the basis for detecting the arc fault, and can only show a high fault identification accuracy rate under certain load conditions. In real residential and factory electrical environments with multiple working conditions, the false alarm rate is very high, which is mainly manifested in two aspects:

[0003] First, there is a problem of missed detection. When a fault arc has occurred, the model detects it as normal, which leads to the risk of electrical fire.

[0004] Second, there is the problem of false picking. Clearly no arc fault has occurred, but the model detects it as an arc fault, causing the arc detector to frequently emit misleading buzzing alarms, which can cause non-fault power outages in sensitive electrical systems and affect production. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a multi-modal fusion fault arc detection method, computer equipment and system, which can effectively reduce the false alarm rate of the arc recognition model and improve the reliability of the arc fault detection system.

[0006] In order to solve the above technical problems, the present invention provides a multi-modal fusion fault arc detection method, including: obtaining multiple groups of modal information under target working conditions; preprocessing the multiple groups of modal information respectively to extract multiple groups of modal feature information; generating multiple groups of modal fault identification results according to the multiple groups of modal feature information respectively, and the modal information, modal feature information and modal fault identification results correspond one to one; fusing the multiple groups of modal fault identification results into a target fault identification result to determine whether there is a fault arc in the target working condition.

[0007] As an improvement of the above-mentioned scheme, the step of generating multiple groups of modal fault identification results according to the multiple groups of modal feature information respectively includes: inputting the multiple groups of modal feature information into the trained multiple groups of modal fault arc identification models respectively to output multiple groups of modal fault identification results; and / or comparing the multiple groups of modal feature information with the preset multiple groups of modal thresholds respectively to output multiple groups of modal fault identification results.

[0008] As an improvement of the above scheme, the step of fusing the multiple groups of modal fault identification results into a target fault identification result to determine whether there is a fault arc in the target working condition includes: dividing the multiple groups of modal fault identification results into a main modal fault identification result and a sub-modal fault identification result; according to the linear conversion formula The weight coefficients of the submodal fault identification results are calculated respectively, where C i is the credibility, W i is a weight coefficient; performing weighted averaging processing on the sub-modal fault identification result according to the weight coefficient to generate an initial fusion result; and generating a target fault identification result according to the initial fusion result and the main modal fault identification result.

[0009] As an improvement of the above-mentioned scheme, the step of calculating the credibility includes: if the accuracy of the modal fault identification result is lower than a preset accuracy, setting the accuracy as the credibility of the modal fault identification result; if the accuracy of the modal fault identification result is equal to the preset accuracy, setting the preset benchmark accuracy as the credibility of the modal fault identification result.

[0010] As an improvement of the above-mentioned scheme, the step of generating a target fault identification result based on the initial fusion result and the main modal fault identification result includes: when the initial fusion result is "fault arc" and the main modal fault identification result is greater than or equal to a preset fault standard, it indicates that a fault arc exists in the target operating condition; when the initial fusion result is "non-fault arc", it indicates that no fault arc exists in the target operating condition.

[0011] As an improvement of the above-mentioned scheme, the multiple groups of modal information include current information, electromagnetic wave information, ultrasonic information and arc information; the multiple groups of modal feature information include current feature information, electromagnetic wave feature information, ultrasonic feature information and arc feature information; the multiple groups of modal fault identification results include current fault identification results, electromagnetic wave fault identification results, ultrasonic fault identification results and arc fault identification results.

[0012] As an improvement of the above-mentioned scheme, the steps of generating multiple groups of modal fault identification results according to the multiple groups of modal feature information respectively include: inputting the current feature information into a trained current fault arc identification model to output a current fault identification result; inputting the electromagnetic wave feature information into a trained electromagnetic wave fault arc identification model to output an electromagnetic wave fault identification result; inputting the ultrasonic feature information into a trained ultrasonic fault arc identification model to output an ultrasonic fault identification result; and comparing the arc feature information with a preset arc threshold to output an arc fault identification result.

[0013] As an improvement of the above scheme, the step of preprocessing the current information to extract current characteristic information includes: format conversion, period segmentation and elimination processing of the current information to form benchmark current information; extracting benchmark characteristic information from the benchmark current information; sorting the benchmark characteristic information by random forest learning method; and combining the benchmark characteristic information into multiple groups of current characteristic information according to the sorting results and preset rules.

[0014] As an improvement of the above-mentioned scheme, the step of preprocessing the electromagnetic wave information to extract electromagnetic wave characteristic information includes: extracting electromagnetic wave main data from the electromagnetic wave information; calculating an electromagnetic wave energy value for every preset number of electromagnetic wave main data; and normalizing and labeling the electromagnetic wave energy values ​​to form electromagnetic wave characteristic information.

[0015] As an improvement of the above scheme, the steps of preprocessing the ultrasonic information to extract ultrasonic characteristic information include: format conversion of the ultrasonic information; FFT Fourier transform of the ultrasonic information after format conversion; extracting positive frequency data from the ultrasonic information after FFT Fourier transform; calculating the ultrasonic energy value based on the positive frequency data; normalizing and labeling the ultrasonic energy value to form ultrasonic characteristic information.

[0016] As an improvement of the above scheme, the step of preprocessing the arc information to extract arc characteristic information includes: converting the arc information into a voltage signal; amplifying the voltage signal; and extracting the amplitude of the amplified voltage signal to form arc characteristic information.

[0017] As an improvement of the above-mentioned scheme, the training steps of the current fault arc identification model include: training K nearest neighbor models through multiple groups of current information under benchmark conditions to form multiple initial current fault arc identification models; testing the accuracy of the initial current fault arc identification models respectively; and taking the initial current fault arc identification model with the highest accuracy as the final current fault arc identification model.

[0018] Correspondingly, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein the processor implements the steps of the above-mentioned multi-modal fusion fault arc detection method when executing the computer program.

[0019] Correspondingly, the present invention also provides a multi-modal fusion fault arc detection system, comprising: a multi-modal information acquisition device, used to collect multiple groups of modal information under target working conditions; the above-mentioned computer device; and a host computer, used to issue a fault warning based on the target fault identification result uploaded by the computer device.

[0020] The implementation of the present invention has the following beneficial effects:

[0021] The present invention introduces multiple groups of modal information, comprehensively considers the original physical phenomenon of the fault arc, integrates the multiple modal information caused by the arc occurrence for analysis, and identifies the fault arc based on multi-dimensional physical characteristics, thereby effectively reducing the false alarm rate of the arc recognition model and improving the reliability of the arc fault detection system;

[0022] Furthermore, the present invention also introduces credibility, weight coefficient, weighted average algorithm and inference rules, etc., to achieve efficient fusion of multi-modal fault identification results with high accuracy and strong applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a flow chart of a first embodiment of a multi-modal fusion fault arc detection method of the present invention;

[0024] Figure 2 is a flow chart of a second embodiment of a multi-modal fusion fault arc detection method of the present invention;

[0025] Figure 3 It is a schematic diagram of the structure of an embodiment of the multi-modal fusion fault arc detection system of the present invention. DETAILED DESCRIPTION

[0026] 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.

[0027] See also Figure 1 , Figure 1 The flowchart of the first embodiment of the multi-modal fusion fault arc detection method of the present invention is shown, which includes:

[0028] S101, obtaining multiple groups of modal information under target working conditions;

[0029] In practical applications, multiple sets of modal information can be collected through sensors of multiple modalities;

[0030] For example, the multiple groups of modal information may be current information, electromagnetic wave information, ultrasonic wave information, arc light information, etc., but are not limited thereto and may be adjusted according to actual conditions.

[0031] S102, preprocessing the multiple groups of modal information respectively to extract multiple groups of modal feature information;

[0032] By preprocessing the modal information, modal feature information in a predefined format can be obtained; wherein the preprocessing can be outlier deletion or replacement, data standardization, missing value filling, filtering, peak detection, distortion processing, data synchronization, etc., but is not limited to these;

[0033] Accordingly, the modal feature information can be saved as a predefined format file to form a data set for subsequent use. Preferably, the file format of the data is csv, xl sx, etc., but it is not limited thereto.

[0034] At the same time, in the present invention, the modal information corresponds one-to-one to the modal characteristic information; for example, multiple groups of modal characteristic information can be current characteristic information, electromagnetic wave characteristic information, ultrasonic characteristic information, arc characteristic information, etc., but it is not limited to this and can be adjusted according to actual conditions.

[0035] S103, generating multiple groups of modal fault identification results according to the multiple groups of modal feature information respectively;

[0036] Similarly, the modal feature information corresponds one-to-one to the modal fault identification results; for example, multiple groups of modal fault identification results include current fault identification results, electromagnetic wave fault identification results, ultrasonic fault identification results and arc fault identification results, but this is not limited to this and can be adjusted according to actual conditions.

[0037] Correspondingly, different fault identification methods can be used for different modal feature information, specifically:

[0038] (1) Inputting the modal feature information into the trained modal fault arc recognition model to output the modal fault recognition result;

[0039] (2) Compare the modal feature information with the preset modal threshold to output the modal fault identification result.

[0040] For example, for current information, the modal fault arc identification model can be used for identification; for electromagnetic wave characteristic information, ultrasonic characteristic information and arc light characteristic information, the modal threshold method can be used for identification; but this is not a limitation and can be adjusted according to actual conditions.

[0041] S104, fusing the multiple groups of modal fault identification results into a target fault identification result to determine whether a fault arc exists in the target working condition.

[0042] Specifically, the steps of fusing multiple groups of modal fault identification results into a target fault identification result to determine whether a fault arc exists in a target working condition include:

[0043] (1) Dividing multiple groups of modal fault identification results into main modal fault identification results and sub-modal fault identification results;

[0044] It should be noted that the modal fault identification result most obviously affected by the fault arc among the multiple groups of modal fault identification results can be used as the main modal fault identification result, and the other modal fault identification results can be used as the secondary modal fault identification results.

[0045] (2) According to the following linear conversion formula, the weight coefficients of the submodal fault identification results are calculated respectively;

[0046]

[0047] Among them, C i is the credibility of the i-th submodal fault identification result, W i is the weight coefficient of the i-th submodal fault identification result.

[0048] Accordingly, the calculation steps of credibility include:

[0049] (1) If the accuracy of the modal fault identification result is lower than the preset accuracy, the accuracy is set as the credibility of the modal fault identification result;

[0050] Generally, the preset accuracy value is 1, but it is not limited to this and can be adjusted according to actual conditions.

[0051] For example, if the accuracy of a modal fault arc identification model is 0.6, the credibility of the corresponding modal fault identification result is set to 0.6.

[0052] For another example, if the accuracy of the modal threshold comparison method is 0.8, the credibility of the corresponding modal fault identification result is set to 0.8.

[0053] (2) If the accuracy of the modal fault identification result is equal to the preset accuracy, the preset benchmark accuracy is set as the credibility of the modal fault identification result.

[0054] Considering the environmental interference problem in practical applications, if the accuracy of the modal fault identification result is equal to the preset accuracy, the preset benchmark accuracy is set as the credibility of the modal fault identification result; among which, the benchmark accuracy is preferably 0.95, but it is not limited to this and can be adjusted according to actual conditions.

[0055] For example, if the accuracy of a modal fault arc identification model is 1, the credibility of the corresponding modal fault identification result is set to 0.95.

[0056] (3) Perform weighted averaging on the submodal fault identification results according to the weight coefficient to generate an initial fusion result;

[0057] For example, if the weighted average result is 0.667, the initial fusion result can be set to "1", indicating "fault arc";

[0058] For example, if the weighted average result is 0.333, the initial fusion result can be set to "0", indicating "non-fault arc".

[0059] (4) Generate target fault identification results based on the initial fusion results and the main modal fault identification results.

[0060] The present invention performs comprehensive judgment based on inference rules to obtain target fault identification results, and the specific steps include:

[0061] (1) When the initial fusion result is "fault arc" and the main modal fault identification result is greater than or equal to the preset fault standard, it means that there is a fault arc in the target working condition;

[0062] Generally, the preset failure standard is 0.9, but it is not limited to this and can be adjusted according to actual conditions.

[0063] (2) When the initial fusion result is “non-fault arc”, it means that there is no fault arc in the target working condition.

[0064] When the initial fusion result is "1", the comprehensive judgment result of all sub-modal fault identification results is considered to be "fault arc". At this time, if the main modal fault identification result is greater than or equal to 0.9, the final result can be determined to be "fault arc";

[0065] When the initial fusion result is "0", the comprehensive judgment result of all sub-modes is determined to be "non-fault arc". Then, no matter what the main mode fault identification result is, the final result is determined to be "non-fault arc".

[0066] In the prior art, the fault arc detection methods are based on a single physical signal as the basis for detecting the fault arc. Different from the prior art, the present invention introduces multiple sets of modal information, comprehensively considers the original physical phenomenon of the fault arc, integrates the multiple modal information caused by the arc occurrence for analysis, and identifies the fault arc based on multi-dimensional physical characteristics, thereby effectively reducing the false alarm rate of the arc recognition model and improving the reliability of the arc fault detection system.

[0067] See also Figure 2 , Figure 2 The flowchart of the second embodiment of the multi-modal fusion fault arc detection method of the present invention is shown; in this embodiment, multiple groups of modal information include current information, electromagnetic wave information, ultrasonic information and arc information, multiple groups of modal feature information include current feature information, electromagnetic wave feature information, ultrasonic feature information and arc feature information, and multiple groups of modal fault identification results include current fault identification results, electromagnetic wave fault identification results, ultrasonic fault identification results and arc fault identification results. The specific process includes:

[0068] S201, obtaining current information, electromagnetic wave information, ultrasonic wave information and arc information under target working conditions;

[0069] The present invention utilizes miniaturized sensor modules such as current sensors, electromagnetic wave sensors, ultrasonic sensors, arc light sensors, etc. installed inside the distribution box to respectively collect corresponding current information, electromagnetic wave information, ultrasonic information, arc light information and other groups of modal information, and sends the above-mentioned multiple groups of modal information to the computer device through wired transmission or wireless transmission.

[0070] S202, preprocessing the current information to extract current characteristic information;

[0071] Accordingly, the step of preprocessing the current information to extract the current characteristic information includes:

[0072] (1) Performing format conversion, period segmentation and elimination processing on the current information to form reference current information;

[0073] Specifically, the current information collected by the oscilloscope can be converted into the JSON format, and the data can be segmented according to the period during the format conversion process, and then the invalid data with the content of "NAN" can be eliminated.

[0074] (2) extracting reference characteristic information from reference current information;

[0075] From the benchmark current information, some time domain features, frequency domain features, and time-frequency features are extracted, including mean, variance, standard deviation, root mean square, skewness, peak-to-peak value, waveform factor, harmonic factor, wavelet detail coefficient, wavelet approximation coefficient and other benchmark feature information.

[0076] (3) Ranking the benchmark feature information through random forest learning method;

[0077] (4) According to the sorting results and preset rules, the baseline characteristic information is combined into multiple groups of current characteristic information.

[0078] The benchmark characteristic information is formed into N groups of current characteristic information in the order and combination of scores from high to low.

[0079] S203, preprocessing the electromagnetic wave information to extract electromagnetic wave characteristic information;

[0080] Accordingly, the step of preprocessing the electromagnetic wave information to extract the electromagnetic wave characteristic information includes:

[0081] (1) extracting electromagnetic wave master data from electromagnetic wave information;

[0082] Delete the header information in the csv file exported by the spectrum analyzer, leaving only the electromagnetic wave main data. The header information is the various parameters set by the acquisition equipment, which is not necessary data for training the model. The electromagnetic wave main data is the useful information of the electromagnetic wave data, including the abnormal change information when the arc fault occurs.

[0083] (2) calculating an electromagnetic wave energy value for each preset number of electromagnetic wave master data;

[0084] The electromagnetic wave master data includes the sampling points and the electromagnetic wave amplitudes corresponding to the points. An electromagnetic wave energy value can be obtained by calculating once for every n points, where n can be 5, 10, 20, 30, etc., but is not limited to this.

[0085] (3) Normalize and label the electromagnetic wave energy values ​​to form electromagnetic wave characteristic information.

[0086] The normalized electromagnetic wave energy characteristic values ​​can be labeled with corresponding labels through a labeling tool; the labeling tool can be a custom python code tool or a public labeling software.

[0087] S204, preprocessing the ultrasonic information to extract ultrasonic feature information;

[0088] Accordingly, the step of preprocessing the ultrasonic information to extract ultrasonic characteristic information includes:

[0089] (1) Convert the format of ultrasonic information;

[0090] In practical applications, the collected xl sx format data can be processed into List format. Specifically, the xl sx format data is read from the target folder, the ultrasonic information is read line by line, and the information such as the load type of the test appliance and the working mode of the appliance is written into a json file and saved in a dictionary format; then, the data processing tool of the programming environment is used to replace the invalid data in the json file based on the mean method, and the processed data is saved in List format.

[0091] (2) performing FFT Fourier transform on the ultrasonic information after format conversion;

[0092] (3) extracting positive frequency data from the ultrasonic information after FFT Fourier transformation;

[0093] (4) Calculating ultrasonic energy value based on positive frequency data;

[0094] (5) Normalize and label the ultrasonic energy values ​​to form ultrasonic characteristic information.

[0095] S205, preprocessing the arc information to extract arc feature information;

[0096] Accordingly, the step of preprocessing the arc information to extract arc characteristic information includes:

[0097] (1) Convert arc information into voltage signals;

[0098] The arc signal is collected by using an optical signal sensor and an arc light collection module, and the arc signal is converted into a voltage signal. At the same time, the voltage signal corresponding to the arc signal is saved.

[0099] (2) Amplifying the voltage signal;

[0100] (3) Extract the amplitude of the amplified voltage signal to form arc characteristic information.

[0101] S206, inputting the current characteristic information into the trained current fault arc recognition model to output a current fault recognition result;

[0102] Accordingly, the training steps of the current fault arc identification model include:

[0103] (1) K nearest neighbor models are trained by using multiple sets of current characteristic information under the benchmark working conditions to form multiple initial current fault arc recognition models;

[0104] It should be noted that the current information under the benchmark operating conditions can be collected, and the current information can be preprocessed to extract the current characteristic information, and then multiple sets of current characteristic information can be used to train the K-nearest neighbor (KNN) machine learning model respectively.

[0105] (2) Test the accuracy of the initial current fault arc identification model respectively;

[0106] (3) The initial current fault arc identification model with the highest accuracy is used as the final current fault arc identification model.

[0107] Finally, the current characteristic information of the target working condition extracted in step S202 is input into the final current fault arc identification model, and the current fault identification result can be output.

[0108] S207, inputting the electromagnetic wave characteristic information into the trained electromagnetic wave fault arc recognition model to output the electromagnetic wave fault recognition result;

[0109] It should be noted that electromagnetic wave information under benchmark working conditions can be collected, and the electromagnetic wave information can be preprocessed to extract electromagnetic wave feature information; then the electromagnetic wave feature information can be divided into training set and test set according to a certain ratio, usually 8:2 or 7:3, with the training set accounting for the majority and the test set accounting for the minority; then, the basic parameters of the machine learning algorithm SVM are set, the training set is used as the input data of the SVM, and the startup function tool of the training model is called to train the electromagnetic wave fault arc recognition model; finally, the test set is used to test the accuracy of the electromagnetic wave fault arc recognition model. If the effect is not ideal, the basic parameters and feature data of the SVM can be continuously adjusted to optimize the electromagnetic wave fault arc recognition model with higher accuracy.

[0110] Finally, the electromagnetic wave characteristic information of the target working condition extracted in step S203 is input into the final electromagnetic wave fault arc identification model, and the electromagnetic wave fault identification result can be output.

[0111] S208, inputting the ultrasonic characteristic information into the trained ultrasonic fault arc recognition model to output the ultrasonic fault recognition result;

[0112] Similarly, ultrasonic information under benchmark working conditions can be collected, and the ultrasonic information can be preprocessed to extract ultrasonic feature information; then the ultrasonic feature information can be divided into training set and test set according to a certain ratio; then, the basic parameters of the machine learning algorithm SVM are set, and the training set is used as the input data of SVM. The startup function tool of the training model is called to train the ultrasonic fault arc recognition model; finally, the test set is used to test the accuracy of the ultrasonic fault arc recognition model. If the effect is not ideal, the basic parameters and feature data of SVM can be continuously adjusted to optimize the ultrasonic fault arc recognition model with higher accuracy.

[0113] Finally, the ultrasonic characteristic information of the target working condition extracted in step S204 is input into the final ultrasonic fault arc recognition model, and the ultrasonic fault recognition result can be output.

[0114] S209, comparing the arc characteristic information with a preset arc threshold to output an arc fault identification result.

[0115] It should be noted that if the arc characteristic information (i.e., amplitude) is greater than the arc threshold, it can be determined that an arc has occurred; then, according to national standards, if an arc occurs for more than 11 half-cycles within 1 second, it can be determined that a "fault arc" has occurred.

[0116] S210, using the current fault identification result as the main modal fault identification result, and using the electromagnetic wave fault identification result, the ultrasonic fault identification result and the arc fault identification result as the sub-modal fault identification result;

[0117] S211, according to the linear conversion formula, respectively calculating the weight coefficients of the electromagnetic wave fault identification result, the ultrasonic fault identification result and the arc fault identification result;

[0118] According to the following linear conversion formula, the weight coefficients of the electromagnetic wave fault identification results, ultrasonic fault identification results and arc fault identification results are calculated in turn:

[0119]

[0120] in,

[0121] C iis the credibility of the i-th submodal fault identification result, that is: C1 is the credibility of the electromagnetic wave fault identification result (that is, the credibility of the electromagnetic wave fault arc identification model), C2 is the credibility of the ultrasonic fault identification result (that is, the credibility of the ultrasonic fault arc identification model), and C3 is the credibility of the arc fault identification result;

[0122] E i is the weight coefficient of the i-th submodal fault identification result, that is: E1 is the weight coefficient of the electromagnetic wave fault identification result, E2 is the weight coefficient of the ultrasonic fault identification result, and W3 is the weight coefficient of the arc fault identification result.

[0123] S212, performing weighted average processing on the electromagnetic wave fault identification result, the ultrasonic fault identification result and the arc fault identification result according to the weight coefficient to generate an initial fusion result;

[0124] A weighted average is taken based on the electromagnetic wave fault identification results, the ultrasonic fault identification results, the arc light fault identification results, the weight coefficients of the electromagnetic wave fault identification results, the weight coefficients of the ultrasonic fault identification results, and the weight coefficients of the arc light fault identification results, and a decision is made based on the weighted average result.

[0125] S213, generating a target fault identification result according to the initial fusion result and the current fault identification result.

[0126] When the initial fusion result in step S212 is "1", the comprehensive determination result of the three signals of electromagnetic wave, ultrasonic wave and arc light is determined to be "fault arc". At this time, if the current fault identification result in step S206 is greater than or equal to 0.9, the final result can be determined to be "fault arc".

[0127] When the initial fusion result in step S212 is "0", the comprehensive judgment result of the three signals of electromagnetic wave, ultrasonic wave and arc light is determined to be "non-fault arc". Then, no matter what the current fault identification result of step S206 is, the final result is determined to be "non-fault arc".

[0128] Therefore, the present invention introduces credibility, weight coefficient, weighted average algorithm and inference rules to achieve efficient fusion of multi-modal fault identification results with high accuracy and strong applicability.

[0129] To summarize, the present invention extracts different modal feature information from each modal information, and uses the modal feature information to train the fault arc recognition model under each modality, and then fuses the fault recognition results of the fault arc recognition model of each modality into the final target fault recognition result based on the strategy fusion method, thereby effectively reducing the false alarm rate of the arc recognition model and improving the reliability of the arc fault detection system.

[0130] See also Figure 3 , Figure 3 The specific embodiment of the multi-modal fusion fault arc detection system 100 of the present invention is shown, which includes a multi-modal information acquisition device 1, a computer device 2 and a host computer 3. The computer device 2 is connected to the multi-modal information acquisition device 1 and the host computer 4 by wired / wireless means. Specifically:

[0131] The multimodal information acquisition device 1 is used to collect multiple groups of modal information under target working conditions and send the multiple groups of modal information to the computer device 2; wherein the multimodal information acquisition device 1 can be a current sensor, an electromagnetic wave sensor, an ultrasonic sensor, an arc light sensor, etc., but is not limited to this.

[0132] A computer device 2, comprising a memory and a processor, wherein the memory stores a computer program, wherein the processor implements the steps of the multi-modal fusion fault arc detection method when executing the computer program, and also uploads the target fault identification result to the host computer 3 through the gateway;

[0133] The host computer 3 is used to issue a fault warning according to the target fault identification result uploaded by the computer device 2, thereby realizing remote fault warning.

[0134] Correspondingly, a terminal system is installed in the host computer 3, wherein the terminal system may be a PC-side electrical safety monitoring system, a WeChat applet on a mobile terminal, etc., but is not limited thereto as long as interaction with the user can be achieved.

[0135] Therefore, the multi-modal fusion fault arc detection system 100 of the present invention can effectively realize remote fault warning and further meet the actual needs of users.

[0136] 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-modal fusion fault arc detection method, characterized in that: include: Obtain multiple sets of modal information under target working conditions; Preprocessing the multiple groups of modal information respectively to extract multiple groups of modal feature information; Generating multiple groups of modal fault identification results according to the multiple groups of modal feature information respectively, wherein the modal information, modal feature information and modal fault identification results correspond one to one; The multiple groups of modal fault identification results are fused into a target fault identification result to determine whether a fault arc exists in the target operating condition.

2. The multi-modal fusion fault arc detection method according to claim 1, characterized in that: The step of generating multiple groups of modal fault identification results according to the multiple groups of modal feature information comprises: inputting the multiple groups of modal feature information into the trained multiple groups of modal fault arc identification models respectively to output multiple groups of modal fault identification results; and / or The multiple groups of modal feature information are compared with multiple groups of preset modal thresholds respectively to output multiple groups of modal fault identification results.

3. The multi-modal fusion fault arc detection method according to claim 1, characterized in that: The step of fusing the multiple groups of modal fault identification results into a target fault identification result to determine whether a fault arc exists in the target working condition comprises: Dividing the multiple groups of modal fault identification results into primary modal fault identification results and secondary modal fault identification results; According to the linear conversion formula The weight coefficients of the submodal fault identification results are calculated respectively, where C i is the credibility, W i is the weight coefficient; Performing weighted averaging processing on the submodal fault identification results according to the weight coefficient to generate an initial fusion result; A target fault identification result is generated according to the initial fusion result and the main mode fault identification result.

4. The multi-modal fusion fault arc detection method according to claim 3, characterized in that: The calculation step of the credibility includes: If the accuracy of the modal fault identification result is lower than a preset accuracy, setting the accuracy as the credibility of the modal fault identification result; If the accuracy of the modal fault identification result is equal to the preset accuracy, the preset reference accuracy is set as the credibility of the modal fault identification result.

5. The multi-modal fusion fault arc detection method according to claim 3, characterized in that: The step of generating a target fault identification result according to the initial fusion result and the main modal fault identification result comprises: When the initial fusion result is "fault arc" and the main modal fault identification result is greater than or equal to the preset fault standard, it means that there is a fault arc in the target working condition; When the initial fusion result is "non-fault arc", it means that there is no fault arc in the target working condition.

6. The multi-modal fusion fault arc detection method according to claim 1, characterized in that: The multiple groups of modal information include current information, electromagnetic wave information, ultrasonic wave information and arc light information; The multiple groups of modal characteristic information include current characteristic information, electromagnetic wave characteristic information, ultrasonic wave characteristic information and arc light characteristic information; The multiple groups of modal fault identification results include current fault identification results, electromagnetic wave fault identification results, ultrasonic fault identification results and arc fault identification results.

7. The multi-modal fusion fault arc detection method according to claim 6, characterized in that: The step of generating multiple groups of modal fault identification results according to the multiple groups of modal feature information comprises: Inputting the current characteristic information into a trained current fault arc recognition model to output a current fault recognition result; Inputting the electromagnetic wave characteristic information into a trained electromagnetic wave fault arc recognition model to output an electromagnetic wave fault recognition result; Inputting the ultrasonic characteristic information into a trained ultrasonic fault arc recognition model to output an ultrasonic fault recognition result; The arc characteristic information is compared with a preset arc threshold to output an arc fault identification result.

8. The multi-modal fusion fault arc detection method according to claim 6, characterized in that: The step of preprocessing the current information to extract current characteristic information includes: Performing format conversion, period segmentation and elimination processing on the current information to form reference current information; extracting reference characteristic information from the reference current information; Sorting the benchmark feature information by random forest learning method; According to the sorting result and preset rules, the reference characteristic information is combined into multiple groups of current characteristic information.

9. The multi-modal fusion fault arc detection method according to claim 6, characterized in that: The step of preprocessing the electromagnetic wave information to extract electromagnetic wave characteristic information comprises: extracting electromagnetic wave master data from the electromagnetic wave information; Calculate an electromagnetic wave energy value for each preset number of electromagnetic wave master data; The electromagnetic wave energy values ​​are normalized and labeled to form electromagnetic wave characteristic information.

10. The multi-modal fusion fault arc detection method according to claim 6, characterized in that: The step of preprocessing the ultrasonic information to extract ultrasonic characteristic information comprises: Converting the ultrasonic information into a format; Perform FFT Fourier transform on the ultrasonic information after format conversion; Extract positive frequency data from the ultrasonic information after FFT Fourier transformation; Calculating ultrasonic energy value according to the positive frequency data; The ultrasonic energy values ​​are normalized and labeled to form ultrasonic characteristic information.

11. The multi-modal fusion fault arc detection method according to claim 6, characterized in that: The step of preprocessing the arc information to extract arc characteristic information comprises: Converting the arc information into a voltage signal; amplifying the voltage signal; The amplitude of the amplified voltage signal is extracted to form arc characteristic information.

12. The multi-modal fusion fault arc detection method according to claim 7, characterized in that: The training steps of the current fault arc identification model include: The K nearest neighbor model is trained by using multiple sets of current information under the benchmark working conditions to form multiple initial current fault arc recognition models; Testing the accuracy of the initial current fault arc identification model respectively; The initial current fault arc identification model with the highest accuracy is taken as the final current fault arc identification model.

13. 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 12 are implemented.

14. A multi-modal fusion fault arc detection system, characterized in that: include: Multimodal information acquisition equipment, used to collect multiple sets of modal information under target working conditions; The computer device of claim 13; The host computer is used to issue a fault warning according to the target fault identification result uploaded by the computer device.

Images