Indoor line series fault arc detection method and system based on multi-feature fusion

Abstract

The invention discloses an indoor line series fault arc detection method based on multi-feature fusion, and the method comprises the steps: collecting a current signal at a main line of an indoor line, carrying out the normalization preprocessing of the current signal, and extracting a time domain, a frequency domain and multiple types of features of the time domain and the frequency domain, so as to comprehensively reflect the non-stationary features of the current signal. For samples under different working conditions, redundant features are removed by using an improved feature selection algorithm in combination with Pearson correlation analysis, and a final feature set with strong discrimination capability is constructed. Furthermore, an extreme learning machine model is introduced, rapid modeling and efficient classification are achieved based on a random initialization weight and a generalized inverse solution method, and high detection precision and real-time performance can be kept under complex working conditions and noise interference. The method has the advantages of being comprehensive in feature expression, efficient in model training, high in detection accuracy and high in robustness, the missing report rate and the false report rate of the indoor line series fault arc can be effectively reduced, and the safety and the reliability of an electrical system are improved.

Description

Technical Field

[0001] This invention relates to the field of electrical engineering technology, specifically to a method and system for detecting arc faults in indoor wiring based on multi-feature fusion. Background Technology

[0002] With the continuous growth of residential electricity load and the widespread use of various electrical appliances, the safe operation of indoor electrical circuits has become increasingly important. Series fault arcing, as one of the main hidden dangers causing electrical fires, is caused by factors such as aging wiring, loose connections, and damaged insulation. In its early stages, the fault often manifests as a weak current distortion, which is difficult to detect using traditional overcurrent protection devices. However, if it continues to develop, it can easily lead to fire accidents, posing a serious threat to life and property. Therefore, developing efficient and accurate indoor circuit series fault arcing detection technology is of significant practical importance.

[0003] Existing arc fault detection technologies can be mainly divided into two categories: identifying arc faults by detecting significant physical phenomena (i.e., non-electrical signals) accompanying the arc occurrence, and identifying arc faults by analyzing the current and voltage distortions caused by the arc using mathematical processing methods. Non-electrical quantity detection methods can directly capture the physical phenomena during arc occurrence, are insensitive to load type, and can complement electrical quantity detection. In specific applications, they help reduce the false alarm and false positive rates. However, they are limited by complex deployment, weak anti-interference capabilities, poor real-time performance, and difficulty in remote intelligent monitoring, making it difficult to meet the practicality and scalability requirements of arc fault detection in modern homes and buildings. Mathematical analysis methods based on electrical waveform distortion can easily achieve online monitoring, with wide coverage and low cost. However, they are prone to false alarms or false negatives in strong interference environments, and their detection performance depends on feature extraction and algorithm design. Existing feature detection methods mainly rely on single-dimensional features, which are easily affected by load characteristics in complex load scenarios (such as multi-electrical appliance operation), leading to false alarms or false negatives. Traditional machine learning algorithms, such as those using support vector machines and decision trees to build detection models, require manual feature selection and extensive parameter tuning. The model's generalization ability is greatly affected by the rationality of feature selection, and the computational complexity is high in high-dimensional feature scenarios, making it difficult to meet real-time detection requirements.

[0004] In addition, indoor wiring loads are complex and multiple appliances often operate simultaneously, with significant differences in current characteristics between different loads, further increasing the difficulty of detecting fault arcs. Summary of the Invention

[0005] To address the shortcomings of current indoor line series arc detection methods, this invention provides an indoor line series fault arc detection method based on multi-feature fusion. This method can adaptively select key features and efficiently train the model, and accurately and quickly identify fault arcs under single-load operation and multiple load combination operation conditions using the indoor line main circuit current signal.

[0006] According to one aspect of the present invention, a method for detecting arc faults in indoor circuits based on multi-feature fusion is provided, comprising: Obtain the current at the main circuit of the indoor wiring and extract the key features of the current. The extracted key features are input into the trained fault arc detection model, and the current state detection result is output; wherein, the training of the fault arc detection model includes: Acquire the normal operating current signal and series fault arc current signal at the main indoor circuit under single load operation and multiple load combination operation conditions; The current signal is preprocessed, and multi-domain statistical features of the normal operating current signal and the series fault arc current signal are extracted to form an original feature set. Then, a feature selection algorithm is used to select key features with discriminative ability from the original feature set to form the final feature set. A model training set is formed by random sampling based on the final feature set; a fault arc detection model is obtained by training the model using the preset extreme learning machine neural network model with the model training set.

[0007] As a further technical solution, the acquisition of normal operating current signals and series fault arc current signals at the indoor main line under single-load operation and multiple load combination operation conditions includes: Resistive loads, inductive loads, and nonlinear loads were selected as typical loads and connected to the indoor circuit of the experiment in different combinations of single, double, triple, and quadruple electrical appliances. In the indoor circuit of the experiment, an arc generator was connected to each branch containing all connected electrical appliances to simulate different series arc generation scenarios. A current transformer is connected to the main line of the indoor circuit in the experiment to collect the main line current signal. For each scenario, the normal operating current and fault arc current are collected for a preset duration and repeated a set number of times to obtain the normal operating current signal and series fault arc current signal under single load operation and multiple load combination operation conditions.

[0008] As a further technical solution, after obtaining the normal operating current signal and the series fault arc current signal under single-load operation and multiple load combination operation conditions, it also includes: Different categories of current data in each scenario are labeled, and the current signal samples are normalized to obtain the original current dataset. For each sample in the original current dataset, its time-domain statistical features, frequency-domain statistical features, and time-frequency-domain statistical features are extracted, and all extracted features constitute the original feature set.

[0009] As a further technical solution, when selecting key features with discriminative ability from the original feature set using a feature selection algorithm, the feature selection algorithm is as follows: , in The current weight value of feature A is represented; x represents the target sample currently randomly selected; k represents the number of nearest neighbor samples of the same or different classes selected; m represents the total number of random samplings. This represents the i-th nearest neighbor sample of the same type that was extracted. Represents the j-th outlier sample extracted from the j-th class. 1 nearest neighbor sample; C represents the total number of sample categories; Let represent the prior probability of the j-th class of samples; Indicates feature A in the sample and The degree of difference between them.

[0010] As a further technical solution, the method also includes: The samples in the original feature set are selected as target samples. The difference between the target sample and its similar and dissimilar neighbor samples in each feature dimension is calculated. Combined with the prior probability of different categories of samples, each feature is weighted and evaluated. The discrimination weight of each feature is iteratively updated. The key features with strong discrimination ability are selected according to the weight ranking to form the final feature set.

[0011] As a further technical solution, a model training set is formed by random sampling based on the final feature set; the model is then trained using the preset extreme learning machine neural network model on the training set, including: Obtain the input vector and corresponding label value for each training sample in the model training set, and construct the input matrix and output vector; Based on randomly initialized input layer weights and biases, an activation function is used to perform feature mapping on the input matrix to generate the hidden layer output matrix. The fault arc detection model is obtained by using the least squares algorithm to solve the output layer weight parameters based on the error between the hidden layer output matrix and the target output, using the generalized inverse algorithm.

[0012] As a further technical solution, the construction of the input matrix and output vector includes: The extreme learning machine neural network model has multiple hidden layer neurons and activation functions, and its matrix expression is as follows: Where H represents the output matrix of the hidden layer. The output weights connecting the hidden layer neurons and the output neurons are represented by T, which represents the expected output matrix.

[0013] As a further technical solution, the current at the main circuit of the indoor line is collected and key features of the current are extracted. The extracted key features are then input into the trained fault arc detection model, including: Based on the current at the main circuit of the indoor line, extract the key features corresponding to the final feature set from the current signal to be detected; The key features are input into the fault arc detection model to determine whether a series fault arc exists and to issue an early warning.

[0014] According to one aspect of the present invention, an indoor circuit series fault arc detection system based on multi-feature fusion is provided, comprising: The data acquisition module is used to obtain the current at the main circuit of the indoor wiring. A fault arc detection module is used to extract key features of the current, input the extracted key features into a trained fault arc detection model, and output current state detection results; wherein, the training of the fault arc detection model includes: Acquire the normal operating current signal and series fault arc current signal at the main indoor circuit under single load operation and multiple load combination operation conditions; The current signal is preprocessed, and multi-domain statistical features of the normal operating current signal and the series fault arc current signal are extracted to form an original feature set. Then, a feature selection algorithm is used to select key features with discriminative ability from the original feature set to form the final feature set. A model training set is formed by random sampling based on the final feature set; a fault arc detection model is obtained by training the model using the preset extreme learning machine neural network model with the model training set.

[0015] According to one aspect of the present invention, a non-transitory computer-readable storage medium is provided, the non-transitory computer-readable storage medium storing computer instructions that cause the computer to execute the described method for detecting arc faults in indoor circuits based on multi-feature fusion.

[0016] The present invention provides a method for detecting arc faults in indoor circuits based on multi-feature fusion. Compared with the prior art, its advantages are as follows: This invention employs a fault arc identification method based on multi-domain feature fusion and an improved feature selection algorithm. It is highly operable and beneficial for improving the accuracy and real-time performance of arc fault detection. Specifically, by collecting indoor current signals and performing normalization preprocessing, and combining multiple features in the time and frequency domains (including zero coefficient, mean, variance, harmonic amplitude, harmonic factor, total harmonic distortion rate, frequency centroid, multi-layer wavelet entropy and permutation entropy, and overall entropy), the multi-dimensional differences of the arc signal are fully explored. Then, a feature selection algorithm is used to sort the multi-domain features, and correlation analysis and redundancy removal strategies are combined to ensure that the selected features are both discriminative and avoid redundant information. Simultaneously, an extreme learning machine neural network model is introduced to achieve efficient modeling and rapid detection.

[0017] Furthermore, when some features are detected to be ineffective due to noise or interference from electrical operating conditions, the constructed fusion feature system can still maintain detection stability by relying on the remaining independent features, effectively reducing false alarms and false negatives, and improving the robustness of the method under complex operating conditions. This solves the problems of insufficient accuracy due to reliance on single features and low training efficiency due to feature redundancy in existing arc detection methods, and the difficulty in balancing real-time performance and robustness due to the long training time or high computational resources required by traditional neural networks or complex classification algorithms. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the process provided in an embodiment of the present invention; Figure 2 This is the experimental load combination state coding diagram provided in the embodiments of the present invention; Figure 3 This is a schematic diagram of the weighting of features in each domain provided in an embodiment of the present invention; Figure 4 This is a training diagram showing the optimal number of hidden layer neurons in a neural network provided in an embodiment of the present invention. Figure 5 The image shows the detection effect of the arc detection model provided in the embodiment of the present invention. Detailed Implementation

[0020] Existing technologies struggle to balance detection accuracy, real-time performance, and scenario adaptability. Therefore, there is an urgent need for a detection method that can integrate multi-dimensional features, adaptively select key features, and efficiently train models to improve the reliability of detecting series fault arcs in indoor circuits.

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form new technical solutions. Such combinations are not bound by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0022] This invention provides a method for detecting arc faults in indoor circuits based on multi-feature fusion. First, the current at the main circuit of the indoor circuit is acquired and key features of the current are extracted. Then, the extracted key features are input into the trained arc fault detection model, and the current state detection result is output.

[0023] Please see Figure 1 In the present invention, an indoor line series fault arc detection method based on multi-feature fusion is provided. The construction and training of the fault arc detection model specifically includes the following steps:

[0024] Normal operating current and series fault arc current signals are collected at the main circuit of the indoor wiring under single-load and multi-load combination operation conditions. Preferably, resistive loads, inductive loads, and nonlinear loads are selected as typical loads and connected to the indoor wiring in different combinations of single, double, triple, and quadruple electrical appliances. In the indoor wiring, an arc generator is connected to each branch containing all connected appliances to simulate different series arc occurrence scenarios. A current transformer is connected at the main circuit of the indoor wiring to collect the main circuit current signal. For each scenario, the normal operating current and fault arc current are collected for a set duration (e.g., 5 seconds), repeated several times (e.g., 5 times), and the data is input to a microprocessor. The microprocessor is used to receive the detected normal operating current signal and fault arc signal and preprocess the signal.

[0025] This embodiment constructs a true-model experimental platform for series fault arcing in indoor circuits. Under a 220V, 50Hz power frequency voltage, a resistive load (kettle), a resistive-inductive load (drill), a nonlinear load (induction cooker), and a dimmer are selected as typical loads. Figure 2 A total of 47 different combinations of operating conditions were sequentially connected to the indoor circuit of the experiment, where 0 indicates that the appliance is in normal operation or not turned on, and 1 indicates that a fault arc has occurred in the branch where the appliance is located. A current transformer was connected to the main circuit of the indoor circuit of the experiment, and the main circuit current value was collected by an oscilloscope at a sampling frequency of 100kHz. For each scenario, the normal operating current and the fault arc current were collected for 5 seconds, and the sampling was repeated 4 times to form a current sample set for different operating conditions.

[0026] Preferably, different categories of current data in each scenario current data are labeled, and the current signal samples are normalized to obtain the original current dataset; for each sample in the original current dataset, 39 different features are extracted, including time-domain statistical features, frequency-domain statistical features, and time-frequency-domain statistical features, and all features constitute the original feature set.

[0027] In this embodiment, different categories of current data in each scenario are labeled, with normal operating state labeled as 0 and fault arc state labeled as 1. Current data over three power frequency cycles is used as one current signal sample. 100 current signal samples are extracted for each different operating condition. These current signal samples are then normalized to obtain the original current dataset. For each sample in the original current dataset, 39 different features are extracted, including time-domain statistical features, frequency-domain statistical features, and time-frequency-domain statistical features. All these features constitute the original feature set. The normalization formula is as follows: , in These are the normalized values ​​from the original current dataset. For the first in the original current dataset The value of each sampling point, This represents the maximum absolute value of the entire current data segment.

[0028] The extracted time-domain statistical features include the zero-dimensional coefficient of the current, the moving average of the current, the moving variance of the current, and the coefficient of variation of the current, specifically: , in The number of current sampling points. The zero-rest threshold is set to 0.05. For the first Current values ​​at each sampling point To determine a piecewise function, when hour, This indicates that the point is in the zero-rest zone. hour, This indicates that the point is not in the zero-rest zone; The mean of all sample points. For the first Current values ​​at each sampling point; The variance of all sample points; For discrete coefficients, The standard deviation for all sample points. This is the mean of all sample points.

[0029] Based on the normalized original current sequence, a discrete Fourier transform is performed to obtain the current spectrum, specifically: , in For length is The discrete-time series of current, It is a complex exponential basis function.

[0030] Based on the obtained current spectrum, the frequency domain statistical features are further extracted, including the amplitudes of the 1st to 15th odd harmonics of the current, the harmonic factor of the current, the total harmonic distortion rate of the current, and the frequency centroid of the current, specifically: , in The number of current sampling points. The fundamental frequency, Sampling frequency, For the first Frequency index corresponding to the subharmonic. For the complex spectrum after DFT transformation at the frequency index The value at; For the first The amplitude of the second harmonic. The amplitude of the fundamental frequency. This represents the upper limit of the harmonic order. For the current signal The actual frequency value of each frequency domain point This is the magnitude of the amplitude at that point.

[0031] The original current sequence samples were decomposed using a 5-level maximum overlap discrete wavelet transform based on the db3 wavelet basis to obtain 5 levels of detail coefficients. ,in The length of each layer of coefficients is the same as the length of the original signal. Based on the detailed coefficients, permutation entropy and wavelet entropy features are extracted respectively. The permutation entropy uses the embedding dimension. With delay parameters Calculations are performed to obtain signal complexity measures at different scales; wavelet entropy is obtained by calculating the energy proportion of each layer of details and substituting it into the Shannon entropy formula, reflecting the relative uncertainty of energy distribution in different frequency bands; the overall entropy is based on the probability distribution of the entire current signal, comprehensively measuring the overall uncertainty and complexity of the signal. This method can simultaneously characterize the multi-scale non-stationary characteristics of current signals, providing time-frequency domain feature support for subsequent arc fault identification, specifically: , in For the embedding dimension, The set of all possible permutations of sequences is . , For the first In the sequence of layer detail coefficients, an arrangement appears. The probability of; For the first Energy of layer detail factor This represents the total energy of 5 layers of detail. For the first The representative value of each amplitude range, The current amplitude falls into the first The probability of each interval. For the amplitude of the current sequence falling into the first... The number of samples in each interval.

[0032] The multi-domain statistical features of the current signal constitute the original feature set. Since the feature selection algorithm needs to consider the proximity relationships between samples of the same class and samples of different classes during the weight update process, an imbalance in the number of samples in each class may cause the distance metric to favor the majority class, thus affecting the fairness and discriminative power of feature selection. Therefore, in the implementation process, the number of samples in each class is balanced. That is, 1600 normal operating current samples and fault arc current samples are selected from the original feature set samples to participate in the feature selection calculation, ensuring the fairness and stability of the feature ranking. Samples in the feature set are selected sequentially as target samples. By calculating the difference between the target sample and its nearest neighbors of the same class and the nearest neighbors of different classes in each feature dimension, and combining the prior probabilities of samples of different classes, each feature is weighted and evaluated. The discriminative weight of each feature is iteratively updated, and the key features with strong discriminative power are selected according to the weight ranking to form the final feature set. The calculation formula for feature selection is as follows: , in Representation of features The current weight value; This represents the currently randomly selected target sample; Indicates the number of nearest neighbor samples of the same or different categories selected; This indicates the total number of random samples. Indicates the first one drawn One similar nearest neighbor sample; Indicates the first one drawn The first class of heterogeneous samples One nearest neighbor sample; Indicates the total number of sample categories; Indicates the first Prior probability of a class of samples; Representation of features In the sample and The degree of difference between them.

[0033] The weights of features within different domains are calculated using the aforementioned feature selection method, and the features in each domain are sorted from largest to smallest according to their weights, resulting in the following: Figure 3 The feature ranking list for each domain is shown. Within each domain, the three features with the highest weight values ​​are retained and merged with features from other domains to form a feature candidate set. The Pearson correlation coefficient matrix is ​​calculated for the features in the candidate set, and the similarity threshold is set to 0.85. Highly redundant features between different domains are removed. Finally, the moving average of the current and the amplitude of the 7th harmonic of the current are selected to form the final feature set. The specific calculation formula is as follows: , in and The first The first feature and the first The sequence of values ​​of a feature across all samples Features With features covariance, Features standard deviation This is the similarity threshold.

[0034] Random sampling is performed on the final feature set, selecting 200 normal operating current feature samples and 200 fault arc current feature samples as the model training set. The input feature vector and its corresponding label vector for each training sample in the model training set are obtained, and an input matrix and output vector are constructed. Based on randomly initialized input layer weights and bias values, the tanh activation function is used to perform feature mapping on the input matrix to generate a hidden layer output matrix. Using the least squares algorithm, based on the error between the hidden layer output matrix and the target output, the output layer weight parameters are solved using the generalized inverse algorithm to obtain the complete fault arc detection model. The neural network model is as follows: , in for dimensional training samples, for Dimensional label samples, The number of hidden layer neurons. To connect the first The weights of each hidden layer neuron and the input neuron For the first Bias of hidden layer neurons For the first The output weights of each hidden layer neuron.

[0035] When constructing the input matrix and output vector of the neural network, for An arbitrary training sample is specifically represented as: , in express Dimensional training samples; express Dimensional labeled samples; the neural network has Hidden layer neurons and activation function , can be represented as: , in Indicates the first The bias of each hidden layer neuron; Indicates the connection of the first The weights of each hidden layer neuron and the input neuron; Indicates the first The output weights of each hidden layer neuron; the matrix expression of the neural network is: , in The output matrix of the hidden layer is expressed by the following formula: , in The output weights connecting the hidden layer neurons and the output neurons are expressed by the following formula: , , in This represents the expected output matrix. Representation matrix The Moore-Penrose generalized inverse matrix.

[0036] To obtain the optimal number of hidden layer neurons, the number of hidden layer neurons in the model was set sequentially from 5 to 70, with the highest average accuracy as the standard. The model achieved the highest accuracy with 46 neurons in the final feature set. Figure 4As shown in the figure. Using the obtained fault arc detection model, after excluding 400 samples from the training set, 1600 samples (800 normal operating current feature samples and 800 fault arc current feature samples) were randomly selected from the remaining 4300 samples in the final feature set as the test set. Twenty tests were conducted for validation, achieving an average accuracy of 98.77%. One randomly selected result is shown in the figure. Figure 5 As shown.

[0037] In practical applications, the current at the main line of the indoor circuit is collected, and only the key features corresponding to the final feature set in the current signal to be detected are extracted; the key features are input into the fault arc detection model to determine whether there is a series fault arc and to issue an early warning.

[0038] Based on the same inventive concept as the foregoing embodiments, this embodiment of the invention also provides an indoor circuit series fault arc detection system based on multi-feature fusion, comprising:

[0039] The data acquisition module is used to obtain the current at the main circuit of the indoor wiring.

[0040] A fault arc detection module is used to extract key features of the current, input the extracted key features into a trained fault arc detection model, and output current state detection results; wherein, the training of the fault arc detection model includes:

[0041] Acquire the normal operating current signal and series fault arc current signal at the main indoor circuit under single load operation and multiple load combination operation conditions;

[0042] The current signal is preprocessed, and multi-domain statistical features of the normal operating current signal and the series fault arc current signal are extracted to form an original feature set. Then, a feature selection algorithm is used to select key features with discriminative ability from the original feature set to form the final feature set.

[0043] A model training set is formed by random sampling based on the final feature set; a fault arc detection model is obtained by training the model using the preset extreme learning machine neural network model with the model training set.

[0044] Preferably, the data acquisition module can be implemented using an indoor series fault arc simulation experimental platform, and the fault arc detection module can be implemented using data processing equipment such as a processor. The execution of the method steps of each module involved in the above system embodiments can be implemented using the corresponding steps in the method embodiments, and will not be elaborated here.

[0045] Based on the same inventive concept as the foregoing embodiments, this embodiment of the invention also provides a non-transitory computer-readable storage medium storing computer instructions that cause the computer to execute the described method for detecting arc faults in indoor circuits based on multi-feature fusion.

[0046] In summary, this invention provides a method for detecting series fault arcs in indoor power lines based on multi-feature fusion. This method collects current signals from the main circuit of the indoor power line and performs normalization preprocessing to extract multiple features in the time and frequency domains, including zero coefficients, mean, variance, harmonic amplitude, harmonic factor, total harmonic distortion rate, frequency centroid, wavelet entropy, permutation entropy, and overall entropy, to comprehensively reflect the non-stationary characteristics of the current signal. For samples under different operating conditions, an improved feature selection algorithm combined with Pearson correlation analysis is used to remove redundant features, constructing a final feature set with strong discriminative power. Furthermore, an extreme learning machine model is introduced, based on random initialization weights and a generalized inverse solution method to achieve rapid modeling and efficient classification, maintaining high detection accuracy and real-time performance under complex operating conditions and noise interference. This invention has the advantages of comprehensive feature representation, efficient model training, high detection accuracy, and strong robustness, effectively reducing the false alarm rate and false alarm rate of series fault arcs in indoor power lines, and improving the safety and reliability of electrical systems.

[0047] The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, product, or apparatus that includes a series of steps or units, not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0048] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting arc faults in indoor circuits based on multi-feature fusion, characterized in that, include: Obtain the current at the main circuit of the indoor wiring and extract the key features of the current. The extracted key features are input into the trained fault arc detection model, and the current state detection result is output; wherein, the training of the fault arc detection model includes: Acquire the normal operating current signal and series fault arc current signal at the main indoor circuit under single load operation and multiple load combination operation conditions; The current signal is preprocessed, and multi-domain statistical features of the normal operating current signal and the series fault arc current signal are extracted to form an original feature set. Then, a feature selection algorithm is used to select key features with discriminative ability from the original feature set to form the final feature set. A model training set is formed by random sampling based on the final feature set; a fault arc detection model is obtained by training the model using the preset extreme learning machine neural network model with the model training set.

2. The method for detecting arc faults in indoor circuits based on multi-feature fusion according to claim 1, characterized in that, The collection of normal operating current signals and series fault arc current signals at the indoor main line under single-load operation and multiple load combination operation conditions includes: Resistive loads, inductive loads, and nonlinear loads were selected as typical loads and connected to the indoor circuit of the experiment in different combinations of single, double, triple, and quadruple electrical appliances. In the indoor circuit of the experiment, an arc generator was connected to each branch containing all connected electrical appliances to simulate different series arc generation scenarios. A current transformer is connected to the main line of the indoor circuit in the experiment to collect the main line current signal. For each scenario, the normal operating current and fault arc current are collected for a preset duration and repeated a set number of times to obtain the normal operating current signal and series fault arc current signal under single load operation and multiple load combination operation conditions.

3. The method for detecting arc faults in indoor circuits based on multi-feature fusion according to claim 2, characterized in that, After obtaining the normal operating current signal and the series fault arc current signal under single load operation and multiple load combination operation conditions, it also includes: Different categories of current data in each scenario are labeled, and the current signal samples are normalized to obtain the original current dataset. For each sample in the original current dataset, its time-domain statistical features, frequency-domain statistical features, and time-frequency-domain statistical features are extracted, and all extracted features constitute the original feature set.

4. The method for detecting arc faults in indoor circuits based on multi-feature fusion according to claim 3, characterized in that, When selecting key features with discriminative ability from the original feature set using a feature selection algorithm, the feature selection algorithm is as follows: ， in The current weight value of feature A is represented; x represents the target sample currently randomly selected; k represents the number of nearest neighbor samples of the same or different classes selected; m represents the total number of random samplings. This represents the i-th nearest neighbor sample of the same type that was extracted. Represents the j-th outlier sample extracted from the j-th class. 1 nearest neighbor sample; C represents the total number of sample categories; Let represent the prior probability of the j-th class of samples; Indicates feature A in the sample and The degree of difference between them.

5. The method for detecting arc faults in indoor circuits based on multi-feature fusion according to claim 4, characterized in that, The method further includes: The samples in the original feature set are selected as target samples. The difference between the target sample and its similar and dissimilar neighbor samples in each feature dimension is calculated. Combined with the prior probability of different categories of samples, each feature is weighted and evaluated. The discrimination weight of each feature is iteratively updated. The key features with strong discrimination ability are selected according to the weight ranking to form the final feature set.

6. The method for detecting arc faults in indoor circuits based on multi-feature fusion according to claim 1, characterized in that, Random sampling is performed based on the final feature set to form a model training set; Training a pre-defined extreme learning machine neural network model using the model's training set includes: Obtain the input vector and corresponding label value for each training sample in the model training set, and construct the input matrix and output vector; Based on randomly initialized input layer weights and biases, an activation function is used to perform feature mapping on the input matrix to generate the hidden layer output matrix. The fault arc detection model is obtained by using the least squares algorithm to solve the output layer weight parameters based on the error between the hidden layer output matrix and the target output, using the generalized inverse algorithm.

7. The method for detecting arc faults in indoor circuits based on multi-feature fusion according to claim 6, characterized in that, The construction of the input matrix and output vector includes: The extreme learning machine neural network model has multiple hidden layer neurons and activation functions, and its matrix expression is as follows: Where H represents the output matrix of the hidden layer. The output weights connecting the hidden layer neurons and the output neurons are represented by T, which represents the expected output matrix.

8. The method for detecting arc faults in indoor circuits based on multi-feature fusion according to claim 1, characterized in that, Collect the current at the main circuit of the indoor wiring and extract the key features of the current. Input the extracted key features into the trained fault arc detection model, including: Based on the current at the main circuit of the indoor line, extract the key features corresponding to the final feature set from the current signal to be detected; The key features are input into the fault arc detection model to determine whether a series fault arc exists and to issue an early warning.

9. A system for detecting arc faults in indoor wiring based on multi-feature fusion, characterized in that, include: The data acquisition module is used to obtain the current at the main circuit of the indoor wiring. A fault arc detection module is used to extract key features of the current, input the extracted key features into a trained fault arc detection model, and output current state detection results; wherein, the training of the fault arc detection model includes: Acquire the normal operating current signal and series fault arc current signal at the main indoor circuit under single load operation and multiple load combination operation conditions; The current signal is preprocessed, and multi-domain statistical features of the normal operating current signal and the series fault arc current signal are extracted to form an original feature set. Then, a feature selection algorithm is used to select key features with discriminative ability from the original feature set to form the final feature set. A model training set is formed by random sampling based on the final feature set; a fault arc detection model is obtained by training the model using the preset extreme learning machine neural network model with the model training set.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that cause the computer to execute the arc detection method for indoor line series faults based on multi-feature fusion as described in any one of claims 1 to 8.

Images