Low-voltage user-side series arc detection method and device based on artificial intelligence

By constructing an AI-based series arc detection model, electrical signals are monitored and processed in real time, solving the problems of misjudgment and missed judgment of series fault arcs in low-voltage distribution networks. This enables timely early warning and accurate identification of fault arcs, improving the power safety of low-voltage users.

CN118091324BActive Publication Date: 2024-11-08JIANGSU YILI ELECTRIC CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410368674.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-03-28
Publication Date
2024-11-08
Estimated Expiration
2044-03-28

AI Technical Summary

Technical Problem

Existing technologies are insufficient for accurately detecting and identifying series fault arcs in low-voltage distribution networks, and there are problems of misjudgment and missed detection. Especially under complex load conditions and external environmental influences, existing methods are not effective under my country's 220V/50Hz voltage and frequency conditions.

Method used

An artificial intelligence-based method for detecting series arcs on the low-voltage user side is adopted. By constructing a series arc hazard early warning model and a multi-scenario low-voltage series arc identification model, electrical signal data is monitored in real time. Feature extraction and judgment are performed using BP neural network and random forest algorithm, and noise is processed by wavelet transform to achieve pre-fault early warning and post-fault identification.

Benefits of technology

It enables timely early warning and accurate identification of fault arcs, reduces the rate of false and false detections, improves the reliability and safety of fault detection, and ensures the power safety of low-voltage users.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118091324B_ABST
    Figure CN118091324B_ABST
Patent Text Reader

Abstract

The present invention discloses a low-voltage user-side series arc detection method and device based on artificial intelligence, including: constructing a series arc hidden danger warning model for pre-fault warning, predicting the probability value of the fault arc hidden danger in the next cycle based on the electrical signal data and development trend at the current moment, judging that there is a potential fault when it exceeds the preset warning threshold, notifying the on-site personnel to perform detection and verification, and visually displaying the warning information; constructing a multi-scenario low-voltage series arc identification model for post-fault identification, extracting arc-related features based on real-time monitored electrical signal data, and inputting the trained multi-scenario low-voltage series arc identification model to further determine the arc type and state. Through the method and device of the present invention, timely prevention and treatment can be carried out before the fault occurs, and the arc type and state can be accurately identified after the fault occurs, effectively improving the sensitivity and reliability of series fault arc perception and the rapidity of personal safety protection.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of low-voltage distribution network fault detection, and in particular to a low-voltage user-side series arc detection method and device based on artificial intelligence. Background Art

[0002] The low-voltage distribution network, at the final stage of the power system, is a critical link connecting the grid to end users. The reliability and safety of its power supply are crucial to the safety of life, property, and quality of life for users. Arc faults frequently occur on the low-voltage user side. These arc faults can cause a dramatic temperature increase in a short period of time, igniting surrounding flammable materials and potentially causing fires. Arc faults can be categorized into three types based on their location: parallel arc faults, series arc faults, and ground arc faults.

[0003] Series arc faults occur on the phase or neutral line, in series with the load. The fault current of a series arc fault is limited by the line load and is generally close to or even less than the normal operating current of the load, making it difficult to detect with existing overcurrent protection. Furthermore, the current waveform of a series arc fault exhibits a certain degree of randomness and is affected by the load, resulting in complex variations. This further complicates the detection and accurate identification of series arc faults.

[0004] Many scholars have conducted extensive research and exploration on series arc fault detection, achieving certain results. However, there are also some shortcomings: Foreign research focuses primarily on single features, but less on multi-feature fusion judgment. This results in inaccurate identification of arc faults and is prone to misjudgments and missed detections. Furthermore, foreign research focuses on 110V / 60Hz voltage and frequency, which is not applicable to my country's 220V / 50Hz. Domestically, there has been considerable theoretical research on methods for extracting characteristic values ​​of arc faults from voltage and current signals, but most of these conclusions are based on theoretical deductions of a single method under a single operating condition. The complex load conditions in actual distribution networks lead to limitations in single methods for arc fault identification. Furthermore, the influence of the external environment on the detection device increases the frequency of misjudgments and missed detections. A comprehensive and complete solution to the hazards of series arc faults is still lacking. Summary of the Invention

[0005] In order to solve the problems of difficult detection of series fault arcs and misjudgment and missed judgment, the present invention proposes a low-voltage user-side series arc detection method and device based on artificial intelligence.

[0006] In order to achieve the above object, the present invention is implemented through the following technical solutions:

[0007] A low-voltage user-side series arc detection method based on artificial intelligence, the method comprising:

[0008] Construct a series arc hazard warning model for pre-fault warning: Real-time monitoring of electrical signal data on the low-voltage user side and input into a trained series arc hazard warning model. Based on the current electrical signal data and development trends, the probability of a fault arc hazard in the next cycle is predicted. When the probability of a fault arc hazard exceeds the preset warning threshold, it is determined that a potential fault exists, and on-site personnel are notified for testing and verification. The warning information is also displayed visually.

[0009] Construct a multi-scenario low-voltage series arc identification model for post-fault identification: Arc-related features are extracted based on real-time monitored electrical signal data, and the trained multi-scenario low-voltage series arc identification model is input to further determine the arc type and status.

[0010] As a preferred solution of the present invention, the construction of a series arc hazard warning model specifically includes:

[0011] Collect electrical signal data before a fault occurs on the low-voltage user side within a historical period and perform data preprocessing to form a fault discrimination data set; the data preprocessing includes noise removal, filtering, and normalization;

[0012] Extracting fault discrimination features from the fault discrimination data set, constructing a fault discrimination feature data set, and dividing the data set into a training set, a test set, and a validation set;

[0013] The series arc hazard warning model is trained using the training set, and the model parameters and structure are adjusted and optimized. The model is tested and verified using the test set and verification set.

[0014] As a preferred solution of the present invention, the construction of a multi-scenario low-voltage series arc identification model specifically includes:

[0015] A low-voltage series arc fault scenario was established in a low-voltage power safety real-world laboratory. An arc generator was connected in series with different types of loads. Electrical signal data from different load circuits was collected as samples to form a sample sequence.

[0016] The sample sequence was preprocessed, noise was filtered out by wavelet transform, and randomly extracted into training set, test set and validation set in a ratio of 7:2:1;

[0017] Arc-related features are extracted from the preprocessed sample sequence through machine learning algorithms, and an index S is constructed to evaluate the sensitivity of arc-related features in a single sample.

[0018] A multi-scenario low-voltage series arc identification model is constructed based on a random forest algorithm. Highly sensitive arc-related features are screened and input into the multi-scenario low-voltage series arc identification model. The model is trained using a supervised learning algorithm, and custom coefficients are constructed to perform multi-objective optimization of the model's hyperparameters.

[0019] The basic evaluator in the multi-scenario low-voltage series arc identification model is a decision tree. Each decision tree outputs a result. According to the principle of minority obeys majority, the results of all decision trees are statistically analyzed to generate the final result of the entire model.

[0020] As a preferred solution of the present invention, data preprocessing is performed on the electrical signal data, including processing the signal high-frequency noise through wavelet transform, and the formula is:

[0021]

[0022] Where, represents the wavelet coefficient; b0 is the sampling rate, j is the value of the discrete wavelet transform discretization power level, k is a constant related to the sampling time interval; ψ() is the wavelet basis function, and x represents the time variable.

[0023] As a preferred solution of the present invention, the series arc hazard warning model is specifically a warning model based on a BP neural network, the hidden layer of the BP neural network adopts the tansig function, and the output h of the hidden layer is l for:

[0024]

[0025] Where, X i Represents the i-th input of the BP neural network, i∈[1,m]; W il is the connection weight from the input layer to the output layer; b l is the bias; l ranges from 1 to n, where n is the number of hidden layer units;

[0026] The output layer uses the purelin function, and the output is:

[0027]

[0028] Where Y r is the output of the BP neural network, r ranges from 0 to 1, Y0 indicates that there is no potential fault, and Y1 indicates that there is a potential fault; W lr is the connection weight from the hidden layer unit to the output layer unit;

[0029] The output of the BP neural network is converted into the arc fault hidden danger probability value P, the formula is:

[0030] As a preferred solution of the present invention, in the low-voltage series arc fault scenario, the acquisition frequency is set to 10 kHz, and each electrical signal data is assigned a unique number, which is obtained by cutting a small fragment of a sliding time window with a width of 200 from the original signal sequence; the sample sequence includes at least eight different types of loads and 17 operating states.

[0031] As a preferred solution of the present invention, the arc-related characteristics include energy spectrum, fundamental frequency component, third harmonic component, fifth harmonic component, fundamental frequency component rate, third harmonic component rate, fifth harmonic component rate, integral, zero-crossing time, variance, kurtosis and Shannon entropy.

[0032] As a preferred solution of the present invention, the arc-related features are extracted from the preprocessed sample sequence by a machine learning algorithm, and the feature extraction formula is:

[0033]

[0034] Where K represents the extracted eigenvalue; β is the fourth-order central moment, D rms is the effective value of the sample sequence D, N is the length of the sample sequence, D y Represents the data of the y-th sampling point in the sample sequence;

[0035] The calculation formula of index S is:

[0036]

[0037] Where G is the total number of categories of arc-related features, p and q represent the p-th and q-th categories of arc-related features, respectively;

[0038] a is the eigenvalue of the arc-related feature belonging to the qth category in a single sample I, and S(a) represents the sensitivity of the evaluated eigenvalue a; is the average value of all characteristic values ​​in the qth type of arc related characteristics; C p 、C q They represent the number of samples in the range [a-0.5v, a-0.5v] in the p-th and q-th arc-related features respectively;

[0039] v=min(max(v q )-min(v q )), q=0,1,2,…,G-1; max(v q )、min(v q ) represent the maximum and minimum values ​​of the arc-related characteristics respectively;

[0040] The closer S(a) is to 1, the higher the sensitivity is, and the closer S(a) is to 0, the lower the sensitivity is. The characteristic values ​​of S(a) exceeding the preset sensitivity threshold are screened as high-sensitivity arc-related features and input into the multi-scenario low-voltage series arc identification model.

[0041] As a preferred solution of the present invention, the model hyperparameters include: the number of decision trees, the maximum depth of the decision tree, the minimum number of leaf node samples, the upper limit of the number of features considered by each decision tree, and the node sample impurity of the decision tree;

[0042] The construction of custom coefficients performs multi-objective optimization on model hyperparameters, specifically:

[0043] The classification accuracy of the multi-scenario low-voltage series arc identification model is recorded as C, the variance of the classification accuracy is V, the classification time is T, and the calculation formula of the custom coefficient f(c) is:

[0044] f(c)=aC+bV+cT;

[0045] Where a, b, and c represent the weight of the classification accuracy C, the variance of the classification accuracy V, and the classification time T, respectively.

[0046] A detection device for a low-voltage user-side series arc detection method based on artificial intelligence, the device comprising:

[0047] The power supply module is used to provide a stable power supply for the entire detection device to ensure the normal operation of each module;

[0048] The original signal processing module is used to collect electrical signal data from the low-voltage user side in real time and perform data preprocessing;

[0049] The signal sampling module is used to standardize the electrical signal data output by the original signal processing module and convert it into a standard sample format for the series arc hazard warning model and the multi-scenario low-voltage series arc identification model;

[0050] An integrated data-driven model calculation and judgment module is used to integrate the series arc hazard warning model and the multi-scenario low-voltage series arc identification model. The model is trained and tested using input samples. The trained model can then be used to provide pre-fault warning or post-fault identification.

[0051] An instruction output module is used to generate an instruction of a fault warning notification or an arc type judgment result according to the output result of the integrated data-driven model calculation and judgment module;

[0052] The action module is used to perform corresponding actions according to the instructions of the instruction output module, including activating an alarm, sending a notification to on-site personnel, or closing related circuits.

[0053] Compared with the existing technology, the beneficial effects of the present invention are: it can carry out timely prevention and treatment before the fault occurs, and accurately identify the arc type and status after the fault occurs, thereby reducing the loss of life and property caused by electrical faults, effectively improving the sensitivity and reliability of fault arc perception and the rapidity of personal safety protection, timely discovering and eliminating safety hazards, avoiding greater harm to the power grid, and ensuring the personal and equipment safety of low-voltage customers during electricity use. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0055] in:

[0056] Figure 1 is a flow chart of the method of the present invention;

[0057] Figure 2 This is a flow chart of a method for constructing a series arc hazard warning model in an embodiment of the present invention;

[0058] Figure 3 This is a flow chart of a method for constructing a multi-scenario low-voltage series arc identification model in an embodiment of the present invention;

[0059] Figure 4 Schematic diagram of a low-voltage user-side series arc detection device based on artificial intelligence in an embodiment of the present invention. DETAILED DESCRIPTION

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of the present invention.

[0061] like Figure 1 FIG. 1 is an embodiment of the present invention, which provides a low-voltage user-side series arc detection method based on artificial intelligence, including:

[0062] S10: Construct a series arc hazard warning model for pre-fault warning: Real-time monitoring of electrical signal data on the low-voltage user side and input of the trained series arc hazard warning model, predicting the probability value of the fault arc hazard in the next cycle based on the current electrical signal data and development trend. When the probability value of the fault arc hazard exceeds the preset warning threshold, it is judged that there is a potential fault, and the on-site personnel are notified to conduct detection and verification, and the warning information is displayed visually at the same time; this helps to detect potential faults early and take corresponding measures to prevent accidents.

[0063] S20: Construct a multi-scenario low-voltage series arc identification model for post-fault identification: Extract arc-related features based on real-time monitored electrical signal data, and input the trained multi-scenario low-voltage series arc identification model to further determine the arc type and status;

[0064] Arc types include series arcs of resistive loads, series arcs of inductive loads, series arcs of nonlinear loads, etc. Arc states can be divided into operating states and fault states;

[0065] The established model can accurately identify the current scenario based on input arc-related features. Using machine learning-based methods to identify low-voltage series arcs in multiple scenarios, the type and state of arcs can be more accurately determined, providing a more reliable basis for fault diagnosis. Furthermore, by continuously accumulating and updating data samples, the accuracy and robustness of the model can be further optimized, improving the performance of the identification method.

[0066] like Figure 2 As shown, in one embodiment, building a series arc hazard warning model includes the following steps:

[0067] S101: Collect electrical signal data from the low-voltage user side before a fault occurs in the past, and perform data preprocessing to form a fault identification data set; data preprocessing includes noise removal, filtering, and normalization;

[0068] Furthermore, the high-frequency noise of the signal is processed by wavelet transform, and the formula is:

[0069]

[0070] Where, represents the wavelet coefficient; b0 is the sampling rate, j is the value of the discrete wavelet transform discretization power level, k is a constant related to the sampling time interval; ψ() is the wavelet basis function, and x represents the time variable;

[0071] S102: Extracting fault discrimination features from the fault discrimination dataset, constructing a fault discrimination feature dataset, and dividing it into a training set, a test set, and a validation set;

[0072] By monitoring the line voltage and the voltage across the device, the contact voltage at the point where the electrical connection is loose can be obtained. Just before a series arc occurs, the amplitude of this contact voltage will show an upward trend, accompanied by many nonlinear characteristics. At the same time, the current in the line will show a slight downward trend due to the increase in contact resistance, which can also be used as a fault identification feature.

[0073] S103: training the series arc hazard warning model using the training set, adjusting and optimizing the model parameters and structure, and testing and verifying the model using the test set and validation set;

[0074] The series arc hazard warning model is specifically a warning model based on BP neural network. The hidden layer of BP neural network adopts tansig function. The output h of the hidden layer is l for:

[0075]

[0076] Where, X i Represents the i-th input of the BP neural network, i∈[1,m]; W il is the connection weight from the input layer to the output layer; b l is the bias; l ranges from 1 to n, where n is the number of hidden layer units;

[0077] The output layer uses the purelin function, and the output is:

[0078]

[0079] Where Y r is the output of the BP neural network, r ranges from 0 to 1, Y0 indicates that there is no potential fault, and Y1 indicates that there is a potential fault; W lr is the connection weight from the hidden layer unit to the output layer unit;

[0080] The output of the BP neural network is converted into the arc fault hidden danger probability value P, the formula is:

[0081] like Figure 3 As shown, in another embodiment, constructing a multi-scenario low-voltage series arc identification model specifically includes:

[0082] S201: A low-voltage series arc fault scenario is established in a low-voltage power safety real-world laboratory. An arc generator is connected in series with different types of loads. Electrical signal data from different load circuits is collected as samples to form a sample sequence.

[0083] In a specific embodiment, the acquisition frequency is set to 10 kHz, and each electrical signal data is assigned a unique number, which is obtained by cutting a small segment of a sliding time window with a width of 200 from the original signal sequence; the sample sequence includes at least eight different types of loads and 17 operating states, such as fans, hair dryers, resistance boxes, electronic switches, etc.

[0084] S202: Preprocess the sample sequence, filter out noise through wavelet transform, and randomly extract it into training set, test set, and validation set in a ratio of 7:2:1;

[0085] S203: extracting arc-related features from the preprocessed sample sequence using a machine learning algorithm, and constructing an index S to evaluate the sensitivity of the arc-related features in a single sample;

[0086] Arc-related characteristics include energy spectrum, fundamental frequency component, third harmonic component, fifth harmonic component, fundamental frequency component rate, third harmonic component rate, fifth harmonic component rate, integral, zero-crossing time, variance, kurtosis, Shannon entropy, etc.

[0087] In one embodiment, the feature extraction formula is:

[0088]

[0089] Where K represents the extracted eigenvalue; β is the fourth-order central moment, D rms is the effective value of the sample sequence D, N is the length of the sample sequence, D y Represents the data of the y-th sampling point in the sample sequence;

[0090] The calculation formula of index S is:

[0091]

[0092] Where G is the total number of categories of arc-related features, p and q represent the p-th and q-th categories of arc-related features, respectively;

[0093] a is the eigenvalue of the arc-related feature belonging to the qth category in a single sample I, and S(a) represents the sensitivity of the evaluated eigenvalue a; is the average value of all characteristic values ​​in the qth type of arc related characteristics; C p 、C q They represent the number of samples in the range [a-0.5v, a-0.5v] in the p-th and q-th arc-related features respectively;

[0094] v=min(max(v q )-min(v q )), q=0,1,2,…,G-1; max(v q)、min(v q ) represent the maximum and minimum values ​​of the arc-related characteristics respectively;

[0095] The closer S(a) is to 1, the higher the sensitivity is, and the closer S(a) is to 0, the lower the sensitivity is. The characteristic values ​​of S(a) exceeding the preset sensitivity threshold are selected as high-sensitivity arc-related features;

[0096] S204: Build a multi-scenario low-voltage series arc identification model based on the random forest algorithm. Filter highly sensitive arc-related features and input them into the multi-scenario low-voltage series arc identification model. Use a supervised learning algorithm for training and construct custom coefficients to perform multi-objective optimization of the model's hyperparameters.

[0097] Furthermore, model hyperparameters include: the number of decision trees, the maximum depth of the decision tree, the minimum number of samples for leaf nodes, the upper limit of the number of features considered by each decision tree, the node sample impurity of the decision tree, etc.

[0098] Construct custom coefficients to perform multi-objective optimization on model hyperparameters, specifically:

[0099] The classification accuracy of the multi-scenario low-voltage series arc identification model is recorded as C, the variance of the classification accuracy is V, the classification time is T, and the calculation formula of the custom coefficient f(c) is:

[0100] f(c)=aC+bV+cT;

[0101] Where a, b, and c represent the weight of the classification accuracy C, the variance of the classification accuracy V, and the classification time T, respectively;

[0102] The basic evaluator in the multi-scenario low-voltage series arc identification model is a decision tree. Each decision tree outputs a result. According to the principle of minority obeys majority, the results of all decision trees are statistically analyzed to generate the final result of the entire model.

[0103] like Figure 4 FIG. 1 is another embodiment of the present invention, which provides a detection device for a low-voltage user-side series arc detection method based on artificial intelligence, comprising:

[0104] The power supply module is used to provide a stable power supply for the entire detection device to ensure the normal operation of each module;

[0105] The original signal processing module is used to collect electrical signal data from the low-voltage user side in real time and perform data preprocessing;

[0106] The signal sampling module is used to standardize the electrical signal data output by the original signal processing module and convert it into a standard sample format for the series arc hazard warning model and the multi-scenario low-voltage series arc identification model;

[0107] An integrated data-driven model calculation and judgment module is used to integrate the series arc hazard warning model and the multi-scenario low-voltage series arc identification model. The model is trained and tested using input samples. The trained model can then be used to provide pre-fault warning or post-fault identification.

[0108] An instruction output module is used to generate an instruction of a fault warning notification or an arc type judgment result according to the output result of the integrated data-driven model calculation and judgment module;

[0109] The action module is used to perform corresponding actions according to the instructions of the instruction output module, including activating an alarm, sending a notification to on-site personnel, or closing related circuits.

[0110] The principles of the detection device are explained from the perspectives of sensors, signal processing, feature extraction, threshold determination, and reliability. First, sensor selection: Select a sensor suitable for low-voltage series arc detection. Common sensors include current sensors and voltage sensors. These sensors can monitor current and voltage changes before and after an arc is generated, or indirectly identify the presence of an arc by detecting optical signals. Figure 4 The L, N, and PE in the figure represent the live wire, neutral wire, and ground wire, respectively. Next, signal acquisition and processing occurs: The signals captured by the sensor are collected and processed. This may involve steps such as analog-to-digital conversion, filtering, and amplification to ensure signal quality and extract useful features. Next, feature extraction and analysis occurs: By extracting and analyzing features from the collected signals, the presence of an arc can be identified. This may include methods such as frequency domain analysis, time domain analysis, and energy spectrum analysis to capture specific patterns and features in the arc signal. Finally, threshold determination and alarming occurs: Based on pre-set arc identification criteria, the feature extraction results are compared with the threshold and a determination is made. When an arc is detected, the device should be able to issue a corresponding alarm signal, such as a sound or light, to alert the user to a potential fault. Ultimately, timely arc detection and early warning are achieved, effectively protecting the safety of equipment and personnel.

[0111] To sum up, the present invention can carry out timely prevention and treatment before a fault occurs, and accurately identify the arc type and status after a fault occurs, thereby reducing the loss of life and property caused by electrical faults, effectively improving the sensitivity and reliability of fault arc perception and the rapidity of personal safety protection, timely discovering and eliminating safety hazards, avoiding greater harm to the power grid, and ensuring the personal and equipment safety of low-voltage customers during electricity use.

[0112] It should be understood that various parts of the present application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. All or part of the steps of the above embodiment method can be completed by instructing the relevant hardware through a program, which can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0113] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing module, or each unit may exist physically separately, or two or more units may be integrated into a single module. The aforementioned integrated modules may be implemented in the form of hardware or in the form of software functional modules. If the aforementioned integrated modules are implemented in the form of software functional modules and sold or used as independent products, they may also be stored in a computer-readable storage medium. The storage medium may be a read-only memory, a magnetic disk, or an optical disk, etc.

[0114] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should be included within the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A low-voltage user-side series arc detection method based on artificial intelligence, characterized in that: The method comprises: Construct a series arc hazard warning model for pre-fault warning: Real-time monitoring of electrical signal data on the low-voltage user side and input of the trained series arc hazard warning model, predicting the probability value of the fault arc hazard in the next cycle based on the current electrical signal data and development trend. When the fault arc hazard probability value exceeds the preset warning threshold, it is judged as a potential fault, notifying on-site personnel to conduct detection and verification, and visually displaying the warning information; The construction of the series arc hazard warning model specifically includes: Collect electrical signal data before a fault occurs on the low-voltage user side in the historical period, and perform data preprocessing to form a fault discrimination data set; the data preprocessing includes noise removal, filtering and normalization; Extracting fault discrimination features from the fault discrimination data set, constructing a fault discrimination feature data set, and dividing the data set into a training set, a test set, and a validation set; The series arc hazard warning model is trained by the training set, the model parameters and structure are adjusted and optimized, and the model is tested and verified by the test set and verification set; The series arc hazard warning model is specifically a warning model based on a BP neural network. The hidden layer of the BP neural network adopts a tansig function. The output h of the hidden layer is l for: Where, X i represents the i-th input of the BP neural network, i∈[1,m]; W il is the connection weight from the input layer to the output layer; b l is the bias; l ranges from 1 to n, where n is the number of hidden layer units; The output layer uses the purelin function, and the output is: Where Y r is the output of the BP neural network, r ranges from 0 to 1, Y0 indicates that there is no potential fault, and Y1 indicates that there is a potential fault; W lr is the connection weight from the hidden layer unit to the output layer unit; The output of the BP neural network is converted into the arc fault hazard probability value P, the formula is: Construct a multi-scenario low-voltage series arc identification model for post-fault identification: extract arc-related features based on real-time monitored electrical signal data, and input the trained multi-scenario low-voltage series arc identification model to further determine the arc type and status; The construction of a multi-scenario low-voltage series arc identification model specifically includes: A low-voltage series arc fault scenario was built in a low-voltage power safety real laboratory. An arc generator was connected in series with different types of loads to collect electrical signal data under different types of load circuits as samples to form a sample sequence. The sample sequence was preprocessed, noise was filtered out through wavelet transform, and randomly extracted into training set, test set and validation set in a ratio of 7:2:

1. The arc-related features are extracted from the preprocessed sample sequence through machine learning algorithm. The feature extraction formula is: In the formula, K represents the extracted eigenvalue; β is the fourth-order central moment, D rms is the effective value of the sample sequence D, N is the length of the sample sequence, D y Represents the data of the yth sampling point in the sample sequence; The index S is constructed to evaluate the sensitivity of arc-related features in a single sample. The calculation formula of the index S is: Where G is the total number of categories of arc-related features, p and q represent the p-th and q-th arc-related features respectively; a is the feature value of the q-th arc-related feature in a single sample I, and S(a) represents the sensitivity of the evaluated feature value a; is the average value of all characteristic values ​​in the qth type of arc related characteristics; C p , C q They represent the number of samples in the range [a-0.5v, a-0.5v] in the p-th and q-th arc-related features respectively; v=min(max(v q )-min(v q )), q=0,1,2,…,G-1; max(v q )、min(v q ) represent the maximum and minimum values ​​of the arc-related characteristics respectively; The closer S(a) is to 1, the higher the sensitivity is, and the closer S(a) is to 0, the lower the sensitivity is. The characteristic values ​​of S(a) exceeding the preset sensitivity threshold are selected as high-sensitivity arc-related features; A multi-scenario low-voltage series arc identification model is constructed based on a random forest algorithm, highly sensitive arc-related features are screened and input into the multi-scenario low-voltage series arc identification model, a supervised learning algorithm is used for training, and custom coefficients are constructed to perform multi-objective optimization on model hyperparameters; The model hyperparameters include: the number of decision trees, the maximum depth of the decision tree, the minimum number of leaf node samples, the upper limit of the number of features considered by each decision tree, and the node sample impurity of the decision tree; The construction of custom coefficients performs multi-objective optimization on model hyperparameters, specifically: The classification accuracy of the multi-scenario low-voltage series arc identification model is recorded as C, the variance of the classification accuracy is V, the classification time is T, and the calculation formula of the custom coefficient f(c) is: f(c)=aC+bV+cT; In the formula, a, b, and c represent the classification accuracy C, the variance of the classification accuracy V, and the weight of the classification time T respectively; The basic evaluator in the multi-scenario low-voltage series arc identification model is a decision tree. Each decision tree outputs a result. According to the principle of minority obeys majority, the results of all decision trees are statistically analyzed to generate the final result of the entire model.

2. The artificial intelligence-based low-voltage user-side series arc detection method according to claim 1 is characterized in that: The electrical signal data is preprocessed, including processing the high-frequency noise of the signal through wavelet transform, and the formula is: In the formula, represents the wavelet coefficient; b0 is the sampling rate, j is the value of the discrete wavelet transform discretization power level, k is a constant related to the sampling time interval; ψ() is the wavelet basis function, and x represents the time variable.

3. The artificial intelligence-based low-voltage user-side series arc detection method according to claim 1 is characterized in that: In the low-voltage series arc fault scenario, the acquisition frequency is set to 10 kHz, and each electrical signal data is assigned a unique number, which is obtained by cutting a small fragment of a sliding time window with a width of 200 from the original signal sequence; the sample sequence includes at least eight different types of loads and 17 operating states.

4. The artificial intelligence-based low-voltage user-side series arc detection method according to claim 1 is characterized in that: The arc-related features include energy spectrum, fundamental frequency component, third harmonic component, fifth harmonic component, fundamental frequency component rate, third harmonic component rate, fifth harmonic component rate, integral, zero-crossing time, variance, kurtosis and Shannon entropy.

5. A detection device based on the artificial intelligence-based low-voltage user-side series arc detection method according to any one of claims 1 to 4, characterized in that: The device comprises: The power supply module is used to provide a stable power supply for the entire detection device to ensure the normal operation of each module; The original signal processing module is used to collect the electrical signal data of the low-voltage user side in real time and perform data preprocessing; The signal sampling module is used to standardize the electrical signal data output by the original signal processing module and convert it into a standard sample format for series arc hazard warning model and multi-scenario low-voltage series arc identification model processing; An integrated data-driven model calculation and judgment module is used to integrate the series arc hazard warning model and the multi-scenario low-voltage series arc identification model, and use the input samples to train and test the series arc hazard warning model and the multi-scenario low-voltage series arc identification model respectively, and use the trained series arc hazard warning model and the multi-scenario low-voltage series arc identification model to perform pre-fault warning or post-fault identification; An instruction output module, used to generate an instruction of a fault warning notification or an arc type judgment result according to the output result of the integrated data-driven model calculation judgment module; The action module is used to perform corresponding actions according to the instructions of the instruction output module, including starting the alarm, sending a notification to on-site personnel, or closing related circuits.