Series arc fault detection method in low-voltage power distribution system
By combining decision trees and neural networks in low-voltage distribution systems, a one-dimensional neural tree model is constructed, and using current characteristics and empirical knowledge, the concealment of series arc fault detection and accuracy in small sample environments in low-voltage distribution systems is solved. , realizes high-precision and interpretable arc fault detection.
Patent Information
- Application Number
- CN202411134152.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-19
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-08-19
AI Technical Summary
The existing technology has concealment, randomness and diversity in series arc fault detection in low-voltage distribution systems, which makes detection difficult. Machine learning methods require feature extraction and low accuracy in small sample environments, and the performance of deep learning models is degraded. Lack of explainability.
Combining decision trees and neural networks, a one-dimensional neural tree model is constructed. Through data acquisition, feature extraction and training, the similarity of current flat shoulder, perimeter period and current mutation characteristics are used to build an interpretability model, increase sample weight, and embed empirical knowledge. to improve model interpretability and generalization capabilities.
It realizes high-precision arc fault detection in a small sample environment, simplifies the feature extraction process, improves model stability and interpretability, reduces the amount of model training data, and enhances the dependency and generalization ability of detection.
Smart Images

Figure CN119024113A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of power systems, and more specifically, relates to a method for detecting series arc faults in a low-voltage power distribution system. Background Art
[0002] Series arc faults are one of the main causes of electrical fires in low-voltage power distribution systems. Every year, arc faults cause more than 10,000 home fires, resulting in serious casualties and property losses. Arc faults are usually caused by loose wiring, long-term circuit overload, insulation aging, etc., accompanied by local electrode high-temperature splashing, which in turn ignites nearby combustible materials, causing fire hazards. To prevent potential losses, national electrical regulations include arc fault detection as an emerging early fire detection technology. However, the hidden, random and diverse nature of arc faults makes their detection extremely challenging.
[0003] At present, machine learning and deep learning methods are the two main directions of research on series arc fault detection. Machine learning classification methods represented by decision trees and random forests have high interpretability and show good classification ability in small sample conditions. However, these machine learning-based methods require arc feature extraction before training the model, and their accuracy is inferior to deep learning methods when processing more complex data. Deep learning methods do not require prior feature extraction and have excellent classification capabilities, and have become a research hotspot in this field. Researchers have proposed a variety of arc detection methods based on neural networks, but these studies mainly focus on improving model performance and reducing the number of model parameters and calculations, ignoring the black box characteristics of neural networks. In addition, the performance of deep learning models will decline when the amount of data is insufficient. Therefore, how to combine the arc fault experience summarized in the laboratory into an interpretable model to improve the model's reliability remains an unresolved issue. Summary of the invention
[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a method for detecting series arc faults in a low-voltage power distribution system, which can quickly and accurately identify arc faults by combining a decision tree with a neural network.
[0005] To achieve the above-mentioned object of the invention, the present invention provides a method for detecting a series arc fault in a low voltage power distribution system, characterized in that it comprises the following steps:
[0006] (1) Data collection and preprocessing to obtain current samples of each load;
[0007] (2) Extract the features of the current samples, including shoulders, cycle-to-cycle similarities, and current mutations;
[0008] (3) Build a binary tree consisting of internal nodes and leaf nodes as a one-dimensional neural tree series arc fault detection model;
[0009] (4) Training the one-dimensional neural tree series arc fault detection model until convergence;
[0010] (5) Detect arc faults based on the trained one-dimensional neural tree series arc fault detection model.
[0011] The object of the invention of the present invention is achieved in this way:
[0012] The present invention discloses a method for detecting series arc faults in a low-voltage power distribution system. The method obtains real current data and labels by building a normal and arc data acquisition platform, collects public data sets to verify the generalization ability of a model, summarizes three artificial experiences of flat shoulder, cycle-to-cycle similarity and current mutation by observing current waveforms, and constructs an interpretable model that can be used to process one-dimensional time series data. The prediction results of the combination of the three artificial experiences are compared with the real labels, and the loss weights of inconsistent samples are increased so that the experience can be embedded in the model and assist the model in training. In this way, the problem of training an interpretable neural tree in a small sample environment with fewer labeled samples can be solved by embedding the mechanism, thereby having the advantages of interpretability, good generalization ability, high prediction accuracy, and the like.
[0013] At the same time, the series arc fault detection method of the present invention integrating the experience-based interpretable model also has the following beneficial effects:
[0014] (1) When training traditional machine learning methods such as decision trees and random forests, additional feature extraction steps are required. This process may not only be time-consuming, but may also reduce classification accuracy and cause loss of important information due to improper feature extraction. However, neural trees directly use raw current data for end-to-end training. This method can extract complex and useful features from raw data without human intervention, thereby simplifying the entire process and providing a solid foundation for deployment in embedded devices.
[0015] (2) The arc fault detection method using experience fusion can significantly reduce the amount of data required for model training, making it possible to apply deep learning to arc fault detection in a small sample environment; model learning guided by experience can not only improve classification accuracy, but also show higher stability and generalization ability when processing new data and coping with uncertainty; experience-based models are easier to explain and understand than purely data-driven methods because they rely on known rules and theories.
[0016] (3) Deep neural networks usually lack interpretability, and when processing large or complex data, traditional machine learning methods may lose interpretability due to model complexity; the interpretable neural tree model combines the architecture of the decision tree and the convolutional layer of the neural network, achieving the advantages of shallow model depth, few trainable parameters, and low model complexity; in addition, due to the embedding of the mechanism, the interpretability of this model is further enhanced. In the actual application environment, the model can be fine-tuned according to specific needs, which provides a great guarantee for the reliability of the invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 It is a flow chart of a method for detecting a series arc fault in a low voltage power distribution system of the present invention;
[0018] Figure 2 It is the AC series arc fault experimental circuit diagram;
[0019] Figure 3 It is a schematic diagram of load connection method;
[0020] Figure 4 yes Figure 3 Schematic diagram of current waveform under the load connection mode shown;
[0021] Figure 5 It is a schematic diagram of the training loss of a one-dimensional neural tree series arc fault detection model;
[0022] Figure 6 It is a schematic diagram of the training accuracy of the one-dimensional neural tree series arc fault detection model;
[0023] Figure 7 It is a schematic diagram of the one-dimensional neural tree series arc fault detection model after training;
[0024] Figure 8 This is a diagram of the model training accuracy of 10KHz in the IEEE public dataset;
[0025] Fig. 9 This is a diagram of the model training accuracy of 6400Hz in the IEEE public data set DETAILED DESCRIPTION
[0026] The specific implementation of the present invention is described below in conjunction with the accompanying drawings so that those skilled in the art can better understand the present invention. It should be noted that in the following description, when the detailed description of known functions and designs may dilute the main content of the present invention, these descriptions will be omitted here.
[0027] Example
[0028] Figure 1 The present invention is a flow chart of a method for detecting series arc faults in a low-voltage power distribution system.
[0029] In this embodiment, if Figure 1 As shown, the present invention provides a method for detecting a series arc fault in a low voltage power distribution system, comprising the following steps:
[0030] S1, data collection and preprocessing;
[0031] S1.1. Collect the current signals of the AC lines under different loads in normal and arc fault conditions at equal intervals. Under each load, m current sampling points are collected. The collected current signal is recorded as I = {I1, I2, ..., I j ,…,I h}, I j ={i1,i2,…,i m}, where I j represents the current signal collected under the jth load, i m represents the current value collected at the mth sampling point, and h represents the number of loads;
[0032] In this embodiment, an AC series arc fault test circuit is constructed according to the requirements of the series carbonization path arc test. Figure 2 As shown. The circuit consists of an AC power supply, a load, an arc generator, an air switch, a current transformer and an oscilloscope. The arc generator is selected and designed in full accordance with the standard requirements. Its main structure consists of a carbon fixed electrode, a copper moving electrode, a spiral regulator, etc. There are 18 types of loads in total, including single loads and multiple loads in parallel. The connection methods are as follows Figure 3 As shown in the figure, 18,000 current sampling points are collected for each load, including the current in normal and fault conditions. Figure 4 It is a schematic diagram of normal and fault current waveforms of three different loads at 3000 sampling points;
[0033] S1.2, the current signal I collected under each load j Perform framing;
[0034] Assume that each frame contains w current sampling points, then each load has Frame current sample, record Then the current sample of each load is expressed as: {I j1 ,I j2 ,…,I jl …,I jd}; In this example, each frame of current samples contains 600 sampling points, with a total of 540 samples;
[0035] Then each frame of current samples is labeled with 0 or 1, where 0 indicates normal and 1 indicates arc fault;
[0036] S2, extracting the characteristics of the current sample;
[0037] S2.1. Extract each current sample I jl The shoulder;
[0038] When a series arc fault occurs, the loop current will have a "zero rest" phenomenon near zero current. Compared with the loop current in a normal state, the current shoulder time during an arc fault will be longer. In order to quantify this experience, an effective dual-threshold method is designed to detect the shoulder duration length in the power frequency cycle waveform, as follows:
[0039] Record current sample I jl ={i j1 ,i j2 ,…,i jn ,…,i jw},n∈[1,w], determine each current i jn Whether it meets:
[0040] |i jn |<th1&|i jn+1 |<th1&|i jn+2 |<th1&|i jn+3 |<th1&|i jn+3 -i jn |<th2
[0041]
[0042] th2=7%×I rms
[0043] Among them, I rms Indicates the effective value of current;
[0044] If the above conditions are met, the sampling point i jn As the current sample I jl The shoulder of the current sample I jl All current values in the , and then count the total number of shoulders, recorded as current shoulders N_shoulders;
[0045] S2.2. Extract each current sample I jl similarity between cycles;
[0046] When the load is working normally, the current is relatively stable, and the similarity between cycles is high. However, when an arc fault occurs, due to the existence of high-frequency random characteristics, there will be a certain amount of mutation in the current signal, and the similarity between cycles is low. Therefore, the Pearson correlation coefficient can be used to reflect the autocorrelation between current signal cycles. The specific extraction process is as follows:
[0047] S2.2.1. Set a notch filter whose transfer function H(z) is:
[0048]
[0049] Among them, a is the notch coefficient, z represents the Z domain;
[0050] S2.2.2. Using a notch filter to process the current sample I jl Perform notch filtering to obtain filtered current samples
[0051] S2.2.3, set current sample It contains t = 2 cycles, that is, each cycle has η = w / t sampling points, then the Pearson correlation coefficient between the cth cycle and the c+1th cycle is recorded as r c,c+1 ;
[0052]
[0053] in, It represents the current value of the kth sampling point in the cth cycle after filtering. It represents the current value of the kth sampling point in the c+1th cycle after filtering. and are the current mean values of all sampling points in the cth cycle and the c+1th cycle respectively;
[0054] S2.2.4. Using the Pearson correlation coefficient between cycles to represent the similarity between cycles, the current sample is calculated using the following formula: The total inter-week similarity is r, denoted as inter-week similarity P_correlation;
[0055]
[0056] S2.3, extract current mutation;
[0057] When an arc fault occurs, the harmonic content of the loop current increases significantly. Compared with the loop current in a normal state, the loop current waveform has obvious burrs when an arc fault occurs. The "burr" phenomenon of resistive load is particularly obvious. Therefore, we can extract the current mutation in the following ways, as follows:
[0058] S2.3.1. Remove each current sample I jl The DC component of:
[0059]
[0060] Among them, i jn_dc Indicates i jn The value after removing the DC component;
[0061] S2.3.2. The current sample after removing the DC component is recorded as I jl_dc ={i j1_dc ,i j2_dc ,…,i jn_dc ,…,i jw_dc}; Then the current sample I jl_dc Perform fast Fourier transform to get the spectrum X jl [k]:
[0062]
[0063] S2.3.3, spectrum X jl [k] is normalized:
[0064]
[0065] S2.3.4. Calculate the current sample I jl Total harmonic distortion THD jl :
[0066]
[0067] The total harmonic distortion is recorded as the current mutation Thd_values;
[0068] S3. Build a one-dimensional neural tree series arc fault detection model;
[0069] The one-dimensional neural tree model is a binary tree structure consisting of internal nodes and leaf nodes, where a solver is deployed on each leaf node and a router is deployed on each internal node; a converter is deployed on each decision path of the binary tree;
[0070] In this embodiment, the converters are distributed on each decision path of the binary tree, and each path has one or more converters, which are nonlinear functions used to convert sample data and pass it to the next node; the converters specifically include convolutional layers, pooling layers, and ReLU activation functions, which take the output of the previous module as input and pass it to the next module after conversion;
[0071] Routers are distributed in each internal node of the binary tree, and distribute input data to its left and right child nodes according to routing decisions; routing decisions are based on random sampling based on Bernoulli distribution;
[0072] The solver is distributed in each leaf node of the binary tree. The input of the solver is the data processed by the converter, and the output is the estimation of the conditional distribution. The solver consists of a one-dimensional convolution, a Relu activation function, a maximum pooling, a flattening operation, a Dropout operation, and a fully connected layer;
[0073] S4, training a one-dimensional neural tree series arc fault detection model;
[0074] S4.1. Select a batch of current samples I from the collected current signal I jl The input is fed into the one-dimensional neural tree series arc fault detection model, and then the converter converts each current sample I jl Perform feature extraction as follows:
[0075] For each current sample I jl The feature is extracted through the converter and recorded as characteristic current
[0076] First, the current sample I jl Perform a one-dimensional convolution operation:
[0077]
[0078] Among them, c(n) is the output of the one-dimensional convolution operation, which represents the convolution result at position n, I jl (n+kp) is the value of the input signal at position n+k, ω(k) is the weight of the convolution kernel at position k, p is the padding length, and K is the convolution kernel length;
[0079] Then the convolution result c(n) is passed through the RuLU activation function:
[0080] ReLU(c(n))=max(0,c(n))
[0081] Finally, the maximum pooling layer is used to reduce the amount of data:
[0082]
[0083] Where H is the size of the pooling window, ξ is the step, and h is the position index within the pooling window;
[0084] S4.2. The characteristic current Input to the router, based on the Bernoulli distribution of characteristic current Random sampling is performed to obtain three routing decisions: split, keep, and stop; among them, split means splitting the input data into two left and right child nodes, keep means continuing to keep the input data in the form of a node to generate the next child node, and stop means not continuing to generate child nodes.
[0085] The router makes three routing decisions for the characteristic current. Perform routing output, the output is recorded as r=1, 2, 3 represent three routing decisions: split, keep, and stop respectively;
[0086] S4.3. Output the route Input to the solver to predict the conditional probabilities corresponding to the three routing outputs;
[0087]
[0088] in, Indicates that given input and model parameters Θ, the model predicts the output Y jl The probability distribution of θ = (θ, ψ, φ) is the parameter set of the model, including the router parameters θ, the converter parameters ψ and the solver parameters φ. Represents input The probability of being assigned to the fth leaf node, represents the specific predicted distribution after selecting the fth leaf node, z∈{0,1} L , z f =1 means that leaf node f is used, and L is the total number of leaf nodes;
[0089] The obtained conditional probability distribution The probability corresponding to each category, that is, the probability of generating a 0 or 1 label, and the category with the highest probability is the model prediction result predicted;
[0090] S4.4. Calculate the current sample I jl The probability of being assigned to leaf node f
[0091]
[0092] Among them, P f represents the set of paths from the root node to the leaf node f, represents the decision probability of the router on the path, is a binary relation that is true when a leaf node f is the left subtree of an internal node o;
[0093] S4.5, Update current sample I jl The weight of
[0094] Set the empirical value P according to the current shoulder N_Shoulders, the current cycle similarity P_correlation and the current mutation Thd_values;
[0095] P=(N_shoulders<=th3)&(P_correlation>=th4)&(Thd_values<th5)*(predicted≠0)
[0096] If the current sample I jl The characteristics of satisfy the above empirical value P, then the weight of the current sample is updated to
[0097] S4.6, using the empirically guided negative log-likelihood as the loss function to calculate the loss values under the three routing outputs;
[0098]
[0099] Among them, p(Y|I weight ,Θ) represents the conditional probability of the model predicting the label given the input and parameters, d is the number of samples in a batch, Y={Y j1 ,Y j2 ,…,Y jl ,…,Y jd} represents the predicted output set of a batch of samples, represents the probability that a batch of inputs is assigned to the fth leaf node, that is, the probability of path selection, is the model output Y under the fth leaf node jl probability;
[0100] S4.7, select the routing output with the smallest loss as the final routing output of this batch of current samples;
[0101] S4.8, repeat steps S4.1-S4.7, through repeated iterative training, until the one-dimensional neural tree series arc fault detection model converges;
[0102] S5, series arc fault detection;
[0103] According to step S1, the current signal of the AC line is collected in real time, and then the current sample is extracted after preprocessing, and input into the trained one-dimensional neural tree series arc fault detection model to detect whether an arc fault occurs.
[0104] The training loss, accuracy and binary tree structure are as follows Figure 5 , 6, 7, Figure 5 The dashed line and solid line represent the loss of the training set and the validation set, respectively, and the loss converges to below 0.2. Figure 6 The dashed line and solid line represent the validation set accuracy and the test set accuracy respectively. Figure 5 and Figure 6 It can be seen that the model has converged and has good performance. Figure 7 The binary tree shown has 3 levels, two internal nodes and three leaf nodes, and one or two transformers are used on each decision path.
[0105] In order to prove the universality of the method of the present invention, we also conducted experiments on the public dataset IAED. The experimental results are as follows: Figure 8 , as shown in 9, Figure 8 and Fig. 9 The training accuracy of the model for 10Khz data and 6400Hz data in the IEEE public data set is shown respectively. It can be seen from the figure that the performance of the present invention on the large sample data set is also very good. The parameters of the public data set are shown in Table 1:
[0106]
[0107]
[0108] Table 1
[0109] As shown in Table 1, the public dataset provides two data resolution settings. The data is collected by a 16-bit, 10kHz high-resolution (HR) recorder (independent measurement device) and a 12-bit, 6400Hz low-resolution (LR) sensor (embedded in a real-time identification system). Arc accidents may occur in one branch of a parallel circuit or in the main branch of a series circuit. Three different types of appliance connections are designed, such as Figure 3 The arc characteristics also depend on the type of appliance. Different load types can introduce different current waveforms. We simulated the arc for ten different household appliances. Figure 3 In the combination shown, the HR data part has 15 different loading configurations, and the LR data part has 10 different loading configurations. The series fault arc detection method of the present invention takes into account the detection accuracy of data sets of different sizes.
[0110] In the actual results, as shown in Table 2, the model based on mechanism embedding in the present invention is better than the traditional machine learning method and deep learning method.
[0111] Performance Indicators Decision Tree SVM 1DCNN ANT The present invention Accuracy 0.8889 0.7222 0.8623 0.9375 0.9583 F1 0.8889 0.8026 0.8500 0.9365 0.9574 Recall 0.8889 0.7222 0.8744 0.9474 0.9649 Precision 0.8889 0.7257 0.8567 0.9333 0.9535
[0112] Table 2
[0113] Although the above describes the illustrative specific embodiments of the present invention to facilitate those skilled in the art to understand the present invention, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations using the concept of the present invention are protected.
Claims
1. A method for detecting series arc faults in a low voltage power distribution system, characterized in that: The following steps are involved: (1) Data collection and preprocessing; (1.1) The current signals of the AC lines under different loads in normal and arc fault conditions are collected at equal intervals. Under each load, m current sampling points are collected. The collected current signal is recorded as I = {I1, I2, ..., I j ,…,I h }, I j ={i1,i2,…,i m }, where I j represents the current signal collected under the jth load, i m represents the current value collected at the mth sampling point, and h represents the number of loads; (1.2) The current signal I collected under each load j Perform framing; Assume that each frame contains w current sampling points, then each load has Frame current sample, record Then the current sample of each load is expressed as: {I j1 ,I j2 ,…,I jl …,I jd }; Then each frame of current samples is labeled with 0 or 1, where 0 indicates normal and 1 indicates arc fault; (2) Extract the characteristics of the current sample; (2.1), extract each current sample I jl The shoulder; Record current sample I jl ={i j1 ,i j2 ,…,i jn ,…,i jw },n∈[1,w], determine each current i jn Whether it meets: |i jn |<th1&|i jn+1 |<th1&|i jn+2 |<th1&|i jn+3 |<th1&|i jn+3 -i jn |<th2 th2=7%×I rms Among them, I rms Indicates the effective value of current; If the above conditions are met, the sampling point i jn As the current sample I jl After traversing the current sample I jl All current values in the , and then count the total number of shoulders, recorded as current shoulders N_shoulders; (2.2), extract each current sample I jl similarity between cycles; (2.2.1) Set a notch filter, whose transfer function H(z) is: Among them, a is the notch coefficient, z represents the Z domain; (2.2.2) Using a notch filter to filter the current sample I jl Perform notch filtering to obtain filtered current samples n∈[1,w]; (2.2.3), let the current sample It contains t cycles, that is, each cycle has m = w / t sampling points, then the Pearson correlation coefficient between the cth cycle and the c+1th cycle is recorded as r c,c+1 ; in, It represents the current value of the kth sampling point in the cth cycle after filtering. It represents the current value of the kth sampling point in the c+1th cycle after filtering. and are the current means of all sampling points in the cth cycle and the c+1th cycle respectively; (2.2.4) Using the Pearson correlation coefficient between cycles to represent the similarity between cycles, the current sample is calculated by the following formula: The total inter-week similarity is r, denoted as inter-week similarity P_correlation; (2.3) Extract current mutation; (2.3.1), remove each current sample I jl The DC component of: Among them, i jn_dc Indicates i jn The value after removing the DC component; (2.3.2), the current sample after removing the DC component is Then the current sample Perform fast Fourier transform to get the spectrum X jl [k]: (2.3.3), spectrum X jl [k] is normalized: (2.3.4), calculate the current sample I jl Total harmonic distortion THD jl : The total harmonic distortion is recorded as the current mutation Thd_values; (3) Build a one-dimensional neural tree series arc fault detection model; The one-dimensional neural tree model is a binary tree structure consisting of internal nodes and leaf nodes. A solver is deployed on each leaf node, which consists of a one-dimensional convolution, a Relu activation function, a maximum pooling, a flattening operation, a Dropout operation, and a fully connected layer. A router is deployed on each internal node. A converter is deployed on each decision path of the binary tree, which includes a convolution layer, a pooling layer, and a Relu activation function. (4) Training a one-dimensional neural tree series arc fault detection model; (4.1) Select a batch of current samples I from the collected current signal I jl The input is fed into the one-dimensional neural tree series arc fault detection model, and then the converter converts each current sample I jl Perform feature extraction as follows: For each current sample I jl The feature is extracted through the converter and recorded as characteristic current First, the current sample I jl Perform a one-dimensional convolution operation: Among them, c(n) is the output of the one-dimensional convolution operation, which represents the convolution result at position n, I jl (n+kp) is the value of the input signal at position n+k, ω(k) is the weight of the convolution kernel at position k, p is the padding length, and K is the convolution kernel length; Then the convolution result c(n) is passed through the RuLU activation function: ReLU(c(n))=max(0,c(n)) Finally, the maximum pooling layer is used to reduce the amount of data: Where H is the size of the pooling window, ξ is the step, and h is the position index within the pooling window; (4.2), the characteristic current Input to the router, based on the Bernoulli distribution of characteristic current Perform random sampling and obtain three routing decisions: split, keep, and stop; The router makes three routing decisions for the characteristic current. Perform routing output, the output is recorded as r=1, 2, 3 represent three routing decisions: split, keep, and stop respectively; (4.3) Output the route Input to the solver to predict the conditional probabilities corresponding to the three routing outputs; in, Indicates that given input and model parameters Θ, the model predicts the output Y jl The probability distribution of θ = (θ, ψ, φ) is the parameter set of the model, including the router parameters θ, the converter parameters ψ and the solver parameters φ. Represents input The probability of being assigned to the fth leaf node, represents the specific predicted distribution after selecting the fth leaf node, z∈{0,1} L , z f =1 means that leaf node f is used, and L is the total number of leaf nodes; The obtained conditional probability distribution Corresponding to the probability of each category, the category with the highest probability is the model prediction result predicted; (4.4), calculate the current sample I jl The probability of being assigned to leaf node f Among them, P f represents the set of paths from the root node to the leaf node f, represents the decision probability of the router on the path, is a binary relation that is true when a leaf node f is the left subtree of an internal node o; (4.5), Update current sample I jl The weight of Set the empirical value P according to the current shoulder N_Shoulders, the current cycle similarity P_correlation and the current mutation Thd_values; P=(N_shoulders<=th3)&(P_correlation>=th4)&(Thd_values<th5)*(predicted≠0) If the current sample I jl The characteristics of satisfy the above empirical value P, then the weight of the current sample is updated to (4.6), using the empirically guided negative log-likelihood as the loss function to calculate the loss values under the three routing outputs; Among them, p(Y|I weight ,Θ) represents the conditional probability of the model predicting the label given the input and parameters, d is the number of samples in a batch, Y={Y j1 ,Y j2 ,…,Y jl ,…,Y jd } represents the predicted output set of a batch of samples, represents the probability that a batch of inputs is assigned to the fth leaf node, that is, the probability of path selection, is the model output Y under the fth leaf node jl probability; (4.7) Select the routing output with the smallest loss as the final routing output of this batch of current samples; (4.8), repeat steps (4.1)-(4.7), and repeat iterative training until the one-dimensional neural tree series arc fault detection model converges; (5) Series arc fault detection; According to step (1), the current signal of the AC line is collected in real time, and then the current sample is extracted after preprocessing and input into the trained one-dimensional neural tree series arc fault detection model to detect whether an arc fault occurs.
Citation Information
Patent Citations
Series arc fault detection method based on one-dimensional convolutional neural network
CN111458599A
Series arc fault detection method based on random forest
CN113552447A
Series arc fault identification method and system based on WRFMDA model
CN115879048A
Fault arc detection method based on one-dimensional dilated convolutional neural network
CN116990648A
Low-voltage series arc fault identification method and device based on multi-dimensional features and XGBoost algorithm, and medium
CN117556371A
Cited By
Low-voltage series fault arc AI detection method and device based on intelligent electric meter
CN121030293A
