CN-121995155-A - Indoor line series fault arc detection method and system based on multi-feature fusion

CN121995155ACN 121995155 ACN121995155 ACN 121995155ACN-121995155-A

Abstract

The invention discloses a multi-feature fusion-based indoor line series fault arc detection method, which comprises the steps of collecting current signals at an indoor line trunk, and carrying out normalization pretreatment on the current signal, and extracting multiple types of characteristics of a time domain, a frequency domain and a time-frequency domain so as to comprehensively reflect the non-stationary characteristic of the current signal. Aiming at samples under different working conditions, redundant features are removed by utilizing 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 realized based on a random initialization weight and a generalized inverse solving method, and higher detection precision and instantaneity can be maintained under complex working conditions and noise interference. The invention has the advantages of comprehensive feature expression, high efficiency of model training, high detection accuracy and strong robustness, can effectively reduce the false alarm rate and false alarm rate of the serial fault arc of the indoor circuit, and improves the safety and reliability of an electrical system.

Inventors

DING LI
ZHOU SHAOJIE
LIU ZIHAN
YU ZHENWEI
ZHOU KAIKAI
KONG ZHENGMIN

Assignees

武汉大学

Dates

Publication Date: 20260508
Application Date: 20251229

Claims (10)

1. The method for detecting the indoor line series fault arc based on multi-feature fusion is characterized by comprising the following steps of: acquiring current at an indoor line trunk and extracting key characteristics of the current; Inputting the extracted key features into a trained fault arc detection model, and outputting a current state detection result, wherein the training of the fault arc detection model comprises the following steps: Acquiring a normal working current signal and a series fault arc current signal of an indoor line main circuit under the conditions of single load operation and multiple load combined operation; The current signal is preprocessed, and multi-domain statistical characteristics of a normal working current signal and a series fault arc current signal are respectively extracted to form an original characteristic set; and training by using the model training set by utilizing a preset extreme learning machine neural network model to obtain a fault arc detection model.
2. The method for detecting the series fault arc of the indoor line based on the multi-feature fusion according to claim 1, wherein the step of collecting the normal working current signal and the series fault arc current signal of the indoor line main line under the conditions of single load operation and multiple load combined operation comprises the following steps: selecting a resistive load, a resistive load and a nonlinear load as typical loads, and sequentially connecting the typical loads into an experimental indoor circuit according to different combination modes of a single electric appliance, a double electric appliance, a three electric appliance and a four electric appliance; In the experimental indoor line, arc generators are connected to branches where all connected electrical appliances are located one by one, and different series arc generation scenes are simulated; And accessing a current transformer at the trunk of the experimental indoor line, collecting trunk current signals, collecting normal working current and fault arc current preset time in each scene, and repeating the setting for times to obtain the normal working current signals and the series fault arc current signals under the conditions of single load operation and multiple load combined operation.
3. The method for detecting the series fault arc of the indoor circuit based on the multi-feature fusion according to claim 2, wherein after obtaining the normal working current signal and the series fault arc current signal under the conditions of single load operation and multiple load combined operation, the method further comprises the following steps: labeling different types of current data in each type of scene current data, and carrying out normalization processing on the current signal samples to obtain an original current data set; and extracting time domain statistical features, frequency domain statistical features and time domain statistical features of each sample in the original current data set, wherein all the extracted features form an original feature set.
4. The method for detecting the series fault arc of the indoor line based on the multi-feature fusion according to claim 3, wherein when key features with discrimination capability are screened out from the original feature set through a feature selection algorithm, the feature selection algorithm is as follows: , Wherein the method comprises the steps of The method comprises the steps of (1) representing a current weight value of a feature A, wherein x represents a current randomly selected target sample, k represents the selected number of similar or dissimilar neighbor samples, and m represents the total number of random sampling; Representing the extracted ith similar neighbor sample; Representing the jth of the decimated samples of the jth class C represents the total number of sample categories; representing the prior probability of the j-th class sample; representing feature A in a sample And Degree of difference between them.
5. The method for detecting an indoor line series fault arc based on multi-feature fusion according to claim 4, wherein the method further comprises: Selecting samples in the original feature set as target samples, calculating the difference degree of the target samples and the similar neighbor and heterogeneous neighbor samples in each feature dimension, carrying out weighted evaluation on each feature by combining the prior probability of different types of samples, iteratively updating the discrimination weight of each feature, and selecting key features with stronger discrimination ability according to the weight ordering to form a final feature set.
6. The method for detecting the indoor line series fault arc based on the multi-feature fusion according to claim 1, wherein the method for detecting the indoor line series fault arc based on the multi-feature fusion is characterized by comprising the steps of randomly sampling according to the final feature set to form a model training set, training by using a preset extreme learning machine neural network model and using the model training set, and comprises the following steps: Obtaining an input vector and a corresponding label value of each training sample in a model training set, and constructing an input matrix and an output vector; based on the randomly initialized input layer weight and bias value, performing feature mapping on the input matrix by adopting an activation function to generate a hidden layer output matrix; and solving the output layer weight parameter by utilizing generalized inverse based on the error between the hidden layer output matrix and the target output through a least square algorithm to obtain the fault arc detection model.
7. The method for detecting an indoor line series fault arc based on multi-feature fusion according to claim 6, wherein the constructing the input matrix and the output vector comprises: The extreme learning machine neural network model is provided with a plurality of hidden layer neurons and an activation function, and the matrix expression is that Wherein H represents the output matrix of the hidden layer, Representing the output weights connecting the hidden layer neurons and the output neurons, T representing the desired output matrix.
8. The method for detecting the series fault arc of the indoor line based on the multi-feature fusion according to claim 1, wherein the steps of collecting the current at the trunk of the indoor line and extracting the key features of the current, and inputting the extracted key features into a trained fault arc detection model comprise the following steps: Extracting key features corresponding to the final feature set in a current signal to be detected according to the current at the indoor line trunk; And inputting the key characteristics into the fault arc detection model to judge whether a series fault arc exists and perform early warning.
9. An indoor line series fault arc detection system based on multi-feature fusion, which is characterized by comprising: the data acquisition module is used for acquiring current at the trunk of the indoor line; The fault arc detection module is used for extracting key features of the current, inputting the extracted key features into a trained fault arc detection model and outputting a current state detection result, wherein the training of the fault arc detection model comprises the following steps: Acquiring a normal working current signal and a series fault arc current signal of an indoor line main circuit under the conditions of single load operation and multiple load combined operation; The current signal is preprocessed, and multi-domain statistical characteristics of a normal working current signal and a series fault arc current signal are respectively extracted to form an original characteristic set; and training by using the model training set by utilizing a preset extreme learning machine neural network model to obtain a fault arc detection model.
10. A non-transitory computer readable storage medium storing computer instructions that cause the computer to perform a multi-feature fusion-based indoor line series fault arc detection method of any one of claims 1 to 8.

Description

Indoor line series fault arc detection method and system based on multi-feature fusion Technical Field The invention relates to the technical field of electrical engineering, in particular to an indoor line series fault arc detection method and system based on multi-feature fusion. Background With the continuous increase of domestic electrical loads and the wide application of various electrical devices, the safe operation of indoor electrical circuits is increasingly important. The series fault arc is one of the main hidden dangers for causing electric fire, and the generation reasons of the series fault arc include line aging, joint loosening, insulation damage and the like, and the series fault arc is often displayed as weak current distortion at the initial stage of the fault, is difficult to identify through the traditional overcurrent protection device, and is extremely easy to cause fire accidents once the series fault arc continuously develops, so that serious threat is formed to life and property safety. Therefore, the research and development of the efficient and accurate indoor line series fault arc detection technology has important practical significance. The existing fault arc detection technology can be mainly divided into the identification of the fault arc by detecting the obvious physical phenomenon (namely, non-electrical quantity signal) accompanied by the occurrence of the arc and the analysis of current and voltage distortion caused by the occurrence of the fault arc by adopting a mathematical processing method so as to realize the identification of the fault arc. The non-electric quantity detection method can directly capture the physical phenomenon when the arc occurs, is insensitive to the load type, can complement the electric quantity detection, is beneficial to reducing the omission factor and the erroneous judgment rate in specific application, but is limited by the problems of complex deployment, weak anti-interference capability, poor real-time performance, difficulty in remote intelligent monitoring and the like, and is difficult to meet the requirements of the modern families and buildings on the practicability and the expandability of fault arc detection. The mathematical analysis method based on electric quantity waveform distortion can conveniently realize on-line monitoring, has wide coverage range and low cost, is easy to misjudge or miss-detect in a strong interference environment, and the detection performance depends on feature extraction and algorithm design, and the existing feature detection method mainly depends on single-dimension features, is easy to be interfered by load characteristics under a complex load (such as multi-electric appliance combined operation) scene, and causes misinformation or miss-information. The traditional machine learning algorithm, such as the construction of a detection model by adopting algorithms such as a support vector machine, a decision tree and the like, needs to manually screen features and carry out a large number of parameter tuning, has great influence on generalization capability of the model by feature selection rationality, has high calculation complexity in a high-dimensional feature scene, and is difficult to meet real-time detection requirements. In addition, the indoor circuit load type is complex, the condition that multiple electric appliances are operated simultaneously is common, the current characteristics of different loads are obviously different, and the difficulty of fault arc detection is further increased. Disclosure of Invention In order to solve the defects of the existing indoor line series arc detection method, the invention provides an indoor line series fault arc detection method based on multi-feature fusion, which can adaptively screen key features and train a model with high efficiency, and complete accurate and rapid identification of fault arcs under the conditions of single load operation and multiple load combined operation through an indoor line trunk current signal. According to an aspect of the present invention, there is provided a method for detecting an indoor line series fault arc based on multi-feature fusion, including: acquiring current at an indoor line trunk and extracting key characteristics of the current; Inputting the extracted key features into a trained fault arc detection model, and outputting a current state detection result, wherein the training of the fault arc detection model comprises the following steps: Acquiring a normal working current signal and a series fault arc current signal of an indoor line main circuit under the conditions of single load operation and multiple load combined operation; The current signal is preprocessed, and multi-domain statistical characteristics of a normal working current signal and a series fault arc current signal are respectively extracted to form an original characteristic set; and training by using the model training set by utilizing