Distribution line fault identification and positioning method based on DWT and BP algorithms

Abstract

The invention relates to the field of discrete signal processing, in particular to a distribution line fault identification and positioning method based on DWT and BP algorithms. Comprising the following steps: S1, carrying hardware equipment on different phases of nodes of the same line in a distribution network line, and collecting and recording discrete data information of current, voltage and traveling waves in an overhead line; s2, constructing a network topological graph of a to-be-monitored line according to the hardware equipment mounting points, and performing noise reduction preprocessing on the original data by combining a particle swarm optimization algorithm and a DWT (Discrete Wavelet Transform) algorithm; and S3, performing polynomial fitting and cumulative curve fitting approximate processing on the data after noise reduction processing to obtain continuous curve data after fitting. According to the distribution line fault identification and positioning method based on the DWT algorithm and the BP algorithm, automatic identification and accurate positioning of distribution line faults are achieved by combining a signal processing technology and an intelligent classification algorithm, the accuracy and efficiency of fault detection are improved, and a powerful guarantee is provided for safe and stable operation of a power grid.

Description

Technical Field

[0001] The present invention relates to the field of discrete signal processing, and in particular to a distribution line fault identification and positioning method based on DWT and BP algorithm. Background Art

[0002] With the rapid development of network technology, network security threats have become increasingly complex and diverse. Network attackers are using increasingly covert, multi-level, and multi-source attack methods, which poses a great challenge to network security. At present, network attack methods are becoming increasingly complex and diverse, and traditional security protection methods are difficult to cope with new threats. Existing network kill chain models such as Cyber ​​Kill Chai have been widely used in security research and defense strategies.

[0003] In the existing technology, although there are some fault detection methods based on signal processing, such as Fourier transform, wavelet transform, etc., these methods have limitations when processing nonlinear and non-stationary signals, and it is difficult to effectively extract periodic information. At the same time, most traditional fault location methods rely on empirical formulas and simple algorithms, and the positioning accuracy and reliability need to be improved.

[0004] For example, the invention patent with the patent publication number CN118584249A, entitled "A distribution line fault detection method based on state feature extraction", includes: (1): signal acquisition; (2): judging according to the current, abnormal, alarm; normal, to (3); (3): Fourier transform of the current; (4): calculating the harmonic distortion rate of the current; (5): judging according to the harmonic distortion rate of the current, abnormal, alarm; normal, to (6); (6): calculating the fundamental distortion rate of the current; (7): judging according to the fundamental distortion rate of the current, abnormal, alarm; normal, to (8); (8): calculating the voltage difference between the two ends of the distribution line; (9): judging whether the voltage difference is greater than the reference value, yes, to (10); no, normal; (10): calculating the abnormal current; (11): judging according to the abnormal current, abnormal, alarm. The disadvantage is that there are limitations in processing nonlinear and non-stationary signals, and it is difficult to effectively extract periodic information. At the same time, relying on empirical formulas and simple algorithms, positioning accuracy and reliability need to be improved. Summary of the invention

[0005] In view of the limitations of distribution line fault detection methods in the prior art when processing nonlinear and non-stationary signals, and the problems of insufficient positioning accuracy and reliability, the present invention provides a distribution line fault identification and positioning method based on DWT and BP algorithm. By combining signal processing technology and intelligent classification algorithm, automatic identification and precise positioning of distribution line faults are achieved, the accuracy and efficiency of fault detection are improved, and a strong guarantee is provided for the safe and stable operation of the power grid.

[0006] To achieve the above technical objectives, the present invention provides a technical solution, which is a method for identifying and locating distribution line faults based on DWT and BP algorithm, comprising the following steps: S1, install the hardware equipment on different phases of the nodes of the same line in the distribution network, collect and record the discrete data information of current, voltage and traveling wave in the overhead line, and upload it to the platform database through the MQTT protocol; S2, build the network topology of the line to be monitored according to the hardware device mounting points, and use the particle swarm optimization algorithm and DWT algorithm to pre-process the original data for noise reduction; S3, performing polynomial fitting and cumulative curve fitting approximation processing on the data after noise reduction preprocessing to obtain continuous curve data after fitting; S4, performing multiple truncation processing on the fitted continuous curve data, and analyzing the frequency components in the truncation interval by fast Fourier transform, and extracting the peak information in the frequency domain as the period information; S5, build a BP neural network model, input the preprocessed data into the network for training and classification recognition, judge the fault interval based on the classification results and the line topology map, and use the traveling wave fault ranging algorithm to locate the fault point.

[0007] In this technical solution, hardware equipment is deployed at key nodes of the distribution network line to collect discrete data of current, voltage and traveling wave in all directions, and efficiently upload them to the cloud database with the help of MQTT protocol; then, particle swarm optimization and DWT technology are used to deeply reduce noise and optimize the quality of original data; then, the data is converted into a continuous curve form through the fitting technology of polynomial and cumulative curve, laying the foundation for accurate analysis. On this basis, multiple data truncation is implemented, and fast Fourier transform is used for frequency domain analysis to extract the peak information of fault data, and the fault data is intercepted according to the peak information. Finally, a BP neural network model is constructed, and the fault data is input into the BP network for training to obtain the network model. The new data is classified and processed through the model results, and the fault interval is quickly locked in combination with the line topology map, and the fault point is accurately located based on the traveling wave fault ranging algorithm. The accuracy and efficiency of fault detection are significantly improved, a solid line of defense is built for the safe and stable operation of the power grid, and the continuous and reliable operation of the power system is effectively guaranteed.

[0008] The present invention is further configured as follows: in step S2, the combination of the particle swarm optimization algorithm and the DWT algorithm for denoising the original data includes: denoising by the DWT algorithm, the DWT algorithm decomposing the signal into approximate coefficients and detail coefficients of different scales, the DWT algorithm using discrete wavelet functions and discrete time scales, and realizing signal decomposition by filtering and downsampling operations, decomposing the original signal into a group of high-frequency sub-signals and low-frequency sub-signals; improving the Heaviside projection function in the DWT algorithm, and optimizing the basic Heaviside function using the sparse regularized Heaviside function; It also includes adding a signal-to-noise ratio comparison module based on the PSO algorithm, and iterating the results processed by the improved DWT algorithm again.

[0009] In this technical solution, by adopting the DWT algorithm, the original signal can be accurately decomposed into approximate coefficients and detail coefficients of different scales. By using discrete wavelet functions and discrete time scales, the separation of high-frequency sub-signals and low-frequency sub-signals of the signal is achieved through filtering and downsampling operations. On the basis of the DWT algorithm, the Heaviside projection function is innovatively improved, and the sparse regularized Heaviside function is introduced. The noise reduction effect is enhanced, providing a more accurate data basis for subsequent fault identification and location. A signal-to-noise ratio comparison module is added to the PSO algorithm framework, and the results after processing by the improved DWT algorithm are iteratively processed by the PSO algorithm to obtain the relatively optimal solution during data preprocessing, thereby improving the accuracy of fault identification.

[0010] The present invention is further configured to: perform sparse regularization optimization on the Heaviside projection function in the DWT algorithm to improve the denoising effect on nonlinear and non-smooth data.

[0011] In this technical solution, during the noise reduction process in step S2, the Heaviside projection function in the DWT algorithm is deeply optimized by sparse regularization, so that the algorithm can show better performance when processing nonlinear and non-smooth data, and improve the accuracy and stability of the noise reduction process.

[0012] The present invention is further configured as follows: in step S3, the cumulative curve fitting approximation processing includes: using a polynomial fitting algorithm to perform preliminary fitting on discrete data, performing secondary fitting through a cumulative curve fitting approximation algorithm, comparing fNIRS data features with similar data features collected by the hardware device, utilizing the behavior of the approximate signal curve in a sliding window, obtaining an estimated curve from multiple overlapping windows, and using a weighted logical recombination method to receive a filtered signal to obtain a final fitting curve.

[0013] In this technical solution, a polynomial fitting algorithm is used to perform a preliminary fit on the original discrete data, and then an advanced CCFA algorithm is used for secondary fitting, which further improves the accuracy and stability of the fitting. The fNIRS data features are introduced for comparative analysis with similar data features collected by the hardware device, and the data features are obtained through data comparison and analysis. The behavior of the approximate signal curve in the sliding window is used to cleverly obtain the estimated curve from multiple overlapping windows, and the weighted logical recombination method is used to receive the filtered signal. Through reasonable weight distribution and logical processing, the relevant noise and interference are effectively eliminated, and the final fitting curve is obtained. This fitting curve not only accurately reflects the actual situation of the distribution line fault, but also provides strong data support for subsequent fault location and identification, thereby significantly improving the accuracy and practicality of the entire method.

[0014] The present invention is further configured as follows: between steps S4 and S5, a processing step is also included: the operation in S4 is iterated, and when the peak information in the frequency domain after the fast Fourier transform tends to be consistent, the iterative operation is terminated to obtain a final peak information set.

[0015] In this technical solution, the fast Fourier transform is used to perform frequency domain analysis on the relevant data, and the peak information in the frequency domain is closely monitored. Through continuous iterative operations, the peak information in the frequency domain gradually converges to a consistent state, thereby effectively eliminating accidental errors and noise interference. When the iterative operation reaches the preset termination condition, that is, when the peak information in the frequency domain shows relative consistency, the iteration is terminated to obtain the final peak information set. It not only accurately reflects the characteristics of distribution line faults in the frequency domain, but also provides accurate data support for subsequent fault location and identification, thereby further improving the accuracy and practicality of the entire method.

[0016] The present invention is further configured as follows: in step S4, the multiple truncation processing of the fitted continuous curve data includes: preferentially setting a preliminary truncation threshold, performing fast Fourier transform on the truncated data, analyzing the frequency components of the data in the truncation interval, obtaining peak information in the frequency domain, performing statistical processing on all peak information, obtaining a preliminary peak information set, and iterating the truncation threshold, performing multiple iterations, and obtaining a new peak information set; after multiple iterations, if the peak information obtained after multiple iterations is maintained within a certain range and the frequency of occurrence is greater than a certain threshold, the peak information is defined as periodic information.

[0017] In this technical solution, first, a preliminary truncation threshold is set, and the continuous curve data is truncated for the first time based on it. Fast Fourier transform is applied to the truncated data to analyze the frequency components of the data in each truncation interval, and the peak information in the frequency domain is accurately captured. The peak information set under the truncation threshold is obtained. By updating the truncation threshold, multiple iterations are performed to obtain a new peak information set. After multiple iterations, if the peak information in the peak information set tends to be relatively stable in multiple iterations, and the frequency of occurrence exceeds the preset threshold, the peak information is defined as periodic information. It not only improves the accuracy of fault identification, but also provides strong support for fault location and repair.

[0018] The present invention is further configured as follows: according to the operation in the processing step, the continuous curve data of n cycles are intercepted from the original data in step S3, and sampling processing is performed to obtain discrete waveform data.

[0019] In this technical solution, after step S3, continuous curve data of n cycles are intercepted from the original data. This ensures that the analyzed data segment has complete and representative cycle characteristics, laying a solid foundation for subsequent fault identification. Next, because discrete data is used as input data during BP neural network training, the continuous curve data is converted into discrete waveform data, that is, the continuous curve data is sampled and processed to obtain the final discrete data. These discrete waveform data not only retain the core characteristics of the original data.

[0020] The present invention is further configured as follows: in step S5, building the BP neural network model includes: creating a training set and a test set, inputting them into the BP neural network for iterative training, obtaining the model parameters of the BP neural network and saving them.

[0021] In this technical solution, first, by creating a training set and a test set, both of which are derived from the discrete waveform data that have been preprocessed and feature extracted in the previous steps. The two sets of data sets are input into the constructed BP neural network. The BP neural network continuously adjusts the weights and bias parameters within the network through multiple iterations of training, in order to achieve the best fitting effect and prediction accuracy. After multiple iterations and adjustments, when the performance of the BP neural network on the training set tends to be stable and meets the termination training conditions, the network model parameters at this time are extracted and saved.

[0022] The present invention is further configured as follows: in step S5, it also includes: performing actual test verification on the BP neural network model, and optimizing the neuron parameters, loss function, optimizer, initialization weights and network structure according to the test verification results to obtain the final waveform classification model.

[0023] In this technical solution, not only the BP neural network model is established, but also it is further tested and verified to ensure the practicality and accuracy of the model. According to the feedback results of these test verifications, the neuron parameters, loss function, optimization algorithm, initialization weights and network structure in the periodic information extraction are comprehensively and systematically optimized.

[0024] The present invention is further configured as follows: in step S5, it also includes: when the classification result of the final waveform classification model is a fault defect, defect interval positioning is performed: the defect interval positioning includes: a short-circuit fault interval judgment algorithm and a single-phase grounding fault interval judgment algorithm.

[0025] In this technical solution, when the optimized final waveform classification model is used to identify the presence of a fault defect in the distribution line, the defect interval positioning program is immediately started. This program includes two core algorithms: the short-circuit fault interval judgment algorithm and the single-phase grounding fault interval judgment algorithm. Through the combined use of these two algorithms, the specific location of the fault can be quickly and accurately locked, providing key guidance information for subsequent fault repair work, thereby greatly improving the efficiency and accuracy of distribution line fault handling.

[0026] The beneficial effects of the present invention are as follows: (1) By combining signal processing technology and intelligent classification algorithm, the automatic identification and precise positioning of distribution line faults can be realized, the accuracy and efficiency of fault detection can be improved, and a strong guarantee can be provided for the safe and stable operation of the power grid; (2) For the denoising of nonlinear data, the Heaviside projection function in the original discrete wavelet transform algorithm is added with sparse regularization, so that its processing effect on nonlinear and non-smooth data is improved, and the particle swarm optimization algorithm is combined to perform secondary denoising on the entire discrete data. By nesting the iterative PSO algorithm behind the improved DWT algorithm, that is, by comparing the signal-to-noise ratio of the improved DWT algorithm with the signal-to-noise ratio of the PSO algorithm, it is also possible to judge whether the denoising effect at this time meets the expectations; (3) A BP neural network is created. The classification of discrete point sets does not require the use of complex neural networks. The BP neural network constructed in this scheme is based on a triangular waveform analysis network, and an expansion experiment is carried out based on this. It is preliminarily concluded that the above network model has a good effect on the classification of the discrete data. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a schematic diagram of the terminal module structure of the present invention; Figure 2 It is a working principle diagram of the terminal of the present invention; Figure 3 This is a schematic diagram of the fault interval determined by the traveling wave of the present invention; Figure 4 A schematic diagram of a short-circuit current fault interval judgment of the present invention; Figure 5 It is a schematic diagram of single-ended and double-ended traveling waves of the present invention; Figure 6 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0028] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation method described herein is only an optimal embodiment of the present invention, which is only used to explain the present invention and does not limit the scope of protection of the present invention. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0029] like Figures 1 to 6 As shown, as a first embodiment of the present invention, a method for identifying and locating distribution line faults based on DWT and BP algorithm includes the following steps: S1, install the hardware equipment on different phases of the nodes of the same line in the distribution network line, collect and record the discrete data information of current, voltage and traveling wave in the overhead line, and upload it to the platform database through the MQTT protocol; S2, build the network topology of the line to be monitored according to the hardware device mounting points, and use the particle swarm optimization algorithm and DWT algorithm to pre-process the original data for noise reduction; S3, performing polynomial fitting and cumulative curve fitting approximation processing on the data after noise reduction preprocessing to obtain fitted continuous curve data; S4, performing multiple truncation processing on the fitted continuous curve data, and analyzing the frequency components in the truncation interval by fast Fourier transform, and extracting the peak information in the frequency domain as the period information; S5, build a BP neural network model, input the preprocessed data into the network for training and classification recognition, judge the fault interval based on the classification results and the line topology map, and use the traveling wave fault ranging algorithm to locate the fault point.

[0030] In this technical solution, hardware equipment is deployed at key nodes of the distribution network line to collect detailed data of current, voltage and traveling wave in all directions, and efficiently upload them to the cloud database with the help of MQTT protocol; then, particle swarm optimization and DWT technology are used to deeply reduce noise and optimize the quality of original data; then, through the fitting technology of polynomial and cumulative curve, the data is converted into a continuous curve form, laying the foundation for accurate analysis. On this basis, multiple data truncation is implemented, and fast Fourier transform is used for frequency domain analysis to obtain the peak information of fault data, and the fault data is intercepted according to the peak information. Finally, a BP neural network model is constructed, and the pre-processed fault data is input for network training, and the fault interval is quickly locked in combination with the line topology map, and the fault point is accurately locked by relying on the traveling wave fault ranging algorithm. The accuracy and efficiency of fault detection are significantly improved, a solid line of defense is built for the safe and stable operation of the power grid, and the continuous and reliable operation of the power system is effectively guaranteed.

[0031] It can be understood that the DWT algorithm is a discrete wavelet transform algorithm.

[0032] In one embodiment of the present invention, in step S2, the denoising preprocessing of the original data by combining the particle swarm optimization algorithm and the DWT algorithm includes: denoising by the DWT algorithm, the DWT algorithm decomposing the signal into approximate coefficients and detail coefficients of different scales, the DWT algorithm using discrete wavelet functions and discrete time scales, and realizing signal decomposition by filtering and downsampling operations, decomposing the original signal into a group of high-frequency sub-signals and low-frequency sub-signals; improving the Heaviside projection function in the DWT algorithm, and optimizing the basic Heaviside function using the sparse regularized Heaviside function; It also includes adding a signal-to-noise ratio comparison module based on the PSO algorithm, and using the improved PSO algorithm to iterate multiple times on the results processed by the improved DWT algorithm.

[0033] In this technical solution, by adopting the DWT algorithm, the original signal can be accurately decomposed into approximate coefficients and detail coefficients of different scales, that is, by using discrete wavelet functions and discrete time scales, through filtering and downsampling operations, the separation of high-frequency sub-signals and low-frequency sub-signals of the signal is achieved, thereby improving the denoising effect. On the basis of the DWT algorithm, the Heaviside projection function is innovatively improved, and the sparse regularized Heaviside function is introduced. The denoising effect is further enhanced. A signal-to-noise ratio comparison module is added to the PSO algorithm framework, and the results after processing by the improved DWT algorithm are iteratively processed by the PSO algorithm to obtain the relatively optimal solution during data preprocessing, thereby improving the accuracy of fault identification.

[0034] It can be understood that since the PSO algorithm is based on a group, the initial values ​​of the group size m, inertia weight w, acceleration constants c1 and c2, maximum speed Vmax, maximum algebra Gmax and other initial values ​​of the PSO algorithm are preliminarily set and the algorithm process is executed to obtain the initial results and calculate its signal-to-noise ratio (SNR); and compare it with the denoised SNR result of the improved DWT algorithm. If the difference between the SNR of the improved DWT algorithm and the SNR of the PSO algorithm is greater than a certain threshold, the process is exited; otherwise, the iterative formula is used to iterate the 6 parameters of the PSO algorithm until the SNR meets the above conditions.

[0035] In step S2, it also includes: sparse regularization optimization of the Heaviside projection function in the DWT algorithm to improve the processing effect of nonlinear and non-smooth data. In the noise reduction process of step S2, the Heaviside projection function in the DWT algorithm is deeply optimized by sparse regularization. The algorithm shows more outstanding performance when processing nonlinear and non-smooth data. Through the introduction of sparse regularization, not only the accuracy and stability of the noise reduction process are improved, but also a clearer and more reliable data basis is provided for subsequent fault identification and location.

[0036] In step S3, the cumulative curve fitting approximation processing includes: using a polynomial fitting algorithm to perform preliminary fitting on discrete data, performing secondary fitting through the CCFA algorithm, comparing the fNIRS data features with the similar data features collected by the hardware device, using the behavior of the approximate signal curve in the sliding window, obtaining an estimated curve from multiple overlapping windows, and using a weighted logic recombination method to receive the filtered signal to obtain the final fitting curve. The polynomial fitting algorithm is used to perform preliminary fitting on the original discrete data, and then the advanced CCFA algorithm is used to perform secondary fitting, which further improves the accuracy and stability of the fitting. The fNIRS data features are introduced for comparative analysis with the similar data features collected by the hardware device, and the data features are obtained through data comparison and analysis. The behavior of the approximate signal curve in the sliding window is used to cleverly obtain the estimated curve from multiple overlapping windows, enriching the information content and reliability of the fitting curve. The weighted logic recombination method is used to receive the filtered signal, and the denoising effect of the data is improved through reasonable weight distribution and logic processing, and the final fitting curve is obtained. This fitting curve not only accurately reflects the actual situation of the distribution line fault, but also provides strong data support for subsequent fault location and identification, thereby significantly improving the accuracy and practicality of the entire method.

[0037] It can be understood that the CCFA algorithm is a special curve fitting technique, which is mainly used to reduce the distortion effect caused by the non-stationarity of fNIRS data.

[0038] Between steps S4 and S5, refer to Figure 2 , and also includes: a processing step: iterating the operation in S4, and when the peak information in the frequency domain after the fast Fourier transform tends to be consistent, terminating the iterative operation to obtain a final peak information set.

[0039] It can be understood that the processing steps specifically include: Processing step 1: setting the initial truncation threshold to the termination truncation threshold, gradually reducing the initial truncation threshold (fixed value), repeating the truncation processing operation, and storing the periodic information in a periodic information set until the termination truncation threshold is reached, and the operation is terminated; Processing step 2: processing the obtained periodic information set, that is, performing a k-means clustering operation on it, so as to obtain its cluster center. If the cluster centers are close, the group of data is assumed to be a periodic function, and any value is taken as the period of the group of data. On the contrary, if there are multiple cluster centers far apart, it is considered that the group of data may contain non-periodic components, but it is mainly sorted by the size of the cluster centers, and the three largest cluster centers are taken as possible periods (in a sense, this may indicate that there is a fault in the data). Fast Fourier transform is used to perform frequency domain analysis on the relevant data. Through continuous iterative operations, the method gradually adjusts and optimizes the processing parameters so that the peak information in the frequency domain gradually tends to be consistent, thereby effectively eliminating accidental errors and noise interference, and improving the reliability and stability of the data. When the iterative operation reaches the preset termination condition, that is, when the peak information in the frequency domain shows a high degree of consistency, the method stops iterating and outputs the final peak information set. It not only accurately reflects the characteristics of distribution line faults in the frequency domain, but also provides accurate data support for subsequent fault location and identification, thereby further improving the accuracy and practicality of the entire method.

[0040] In step S4, the multiple truncation processing of the fitted continuous curve data includes: preferentially setting a preliminary truncation threshold, performing fast Fourier transform on the truncated data, analyzing the frequency components of the data in the truncation interval, obtaining the peak information in the frequency domain, and obtaining the peak information set under the truncation threshold. By updating the truncation threshold and iterating the truncation threshold, multiple iterations are performed to obtain a new peak information set; after multiple iterations, if the peak information obtained after multiple iterations tends to be relatively stable and the frequency of occurrence is greater than the preset threshold, these peak information are defined as periodic information. This not only improves the accuracy of fault identification, but also provides strong support for fault location and repair.

[0041] According to the operation in the processing step, the continuous curve data of n cycles are intercepted from the original data in step S3, and the sampling process is performed to obtain discrete waveform data. It ensures that the analyzed data segment has complete and representative cycle characteristics, laying a solid foundation for subsequent fault identification. Next, in order to convert these continuous curve data into a form more suitable for digital processing, these data are further sampled. Through a scientific sampling strategy, the original continuous curve is accurately converted into a series of discrete waveform data points. These discrete waveform data not only retain the core characteristics of the original data, but also greatly simplify the subsequent data processing process and improve the calculation efficiency.

[0042] In step S5, the construction of the BP neural network model includes: creating a training set and a test set, inputting them into the BP neural network for iterative training, obtaining the model parameters of the BP neural network and saving them. First, by creating a training set and a test set, both sets of data sets are derived from the discrete waveform data that have been preprocessed and feature extracted in the previous steps. Input these two sets of data sets into the constructed BP neural network. The BP neural network continuously adjusts the weights and bias parameters inside the network through multiple iterative training in order to achieve the best fitting effect and prediction accuracy. After multiple iterations and adjustments, when the performance of the BP neural network on the training set tends to be stable and meets the termination training conditions, the network model parameters at this time are extracted and saved. These parameters not only contain the structural information of the neural network, but also contain the fault identification and location knowledge obtained through a large amount of data learning, which is the key to the subsequent rapid and accurate identification and location of faults.

[0043] In step S5, it also includes: performing actual test verification on the BP neural network model, and optimizing the neuron parameters, loss function, optimizer, initialization weights and network structure in the periodic information according to the test verification results to obtain the final waveform classification model. Not only the BP neural network model is established, but also actual test verification is further performed on it to ensure the practicality and accuracy of the model. According to the feedback results of these test verifications, the neuron parameters, loss function, optimization algorithm, initialization weights and network structure in the periodic information extraction are comprehensively and systematically optimized.

[0044] In step S5, it also includes: when the classification result of the final waveform classification model is a fault defect, the defect interval positioning is performed: the defect interval positioning includes: a short-circuit fault interval judgment algorithm and a single-phase grounding fault interval judgment algorithm. When the optimized final waveform classification model is used to identify the presence of a fault defect in the distribution line, the defect interval positioning program is immediately started. This program includes two core algorithms: a short-circuit fault interval judgment algorithm and a single-phase grounding fault interval judgment algorithm. Through the combined use of these two algorithms, the specific location of the fault can be quickly and accurately locked, providing key guidance information for subsequent fault repair work, thereby greatly improving the efficiency and accuracy of distribution line fault handling.

[0045] It can be understood that the short-circuit fault interval judgment algorithm is mainly based on the short-circuit current judgment interval and supplemented by the change in the direction of the traveling wave head; the single-phase grounding fault interval judgment algorithm is mainly based on the change in the polarity of the traveling wave and the judgment of the grounding current; in this stage, it will be supplemented by the network topology diagram constructed in step S2.

[0046] It can be understood that the short-circuit fault interval is determined as follows: if a short circuit occurs at a certain point, a large short-circuit current will appear in the direction of the power supply at that node, and no short-circuit current will flow in the load direction of that node. The early fault indicator operates according to this principle to determine the short-circuit fault interval; Figure 4 As shown: If a short circuit fault occurs in the AB phase at point F1, short circuit current will appear in both the AB phases from point F1 to the power supply direction, but there will be no short circuit current in the load direction and the non-fault phase. It should be noted that the interval refers to the line from the monitoring device to the power supply direction. If a device on the main line has a short circuit current, there must be a short circuit point in the load direction (the main line and branches behind are both possible). For single-phase grounding faults, the single traveling wave has transmission and reflection in the line, and the same device may have the following situations: 1. Receive the first traveling wave signal and report it; 2. Only the reflected wave of the first traveling wave signal is received and reported; 3. Receive the first traveling wave signal and report it, and receive the first reflected traveling wave signal and report it; In view of the above situation, the following characteristics should be analyzed: like Figure 3 As shown in the figure, the single-phase grounding fault interval judgment is that the traveling wave generated at the same fault point at a certain moment propagates to both ends, and the signal amplitude decays with distance. The traveling wave signal amplitude should be gradually decayed from near to far equipment on the line, otherwise there may be a situation where a reflected traveling wave is received; when selecting the traveling wave signal, two traveling waves with opposite polarities and the largest signal amplitude should be selected for interval judgment (generally the equipment before and after the fault point), and according to the time difference of the "inflection point" position of the traveling wave, calculate whether it meets the time requirements for signal propagation, otherwise it cannot be used as a condition for fault interval judgment.

[0047] It can be understood that in step S1, the power frequency data (normal operation data) is sent once at a certain interval, while the high frequency data (line fault data) is uploaded in real time; the data formats of the two are consistent, and the formats are as follows: Traveling wave data (data packets in this format are sent 40 times in succession, and the data content contains a total of (6400*40) points); Voltage data (data packets in this format are sent 10 times in succession, and the data content contains a total of (6400*10) points); Current data (data packets in this format are sent 10 times in succession, and the data content contains a total of (6400*10) points).

[0048] It can be understood that in step S5, in the construction of the BP neural network, the BP neural network mainly consists of 1 input layer, 5 hidden layers and 1 output layer, which specifically includes the following steps: Input layer: 1280 neural units; first hidden layer: 1536 neural units; second hidden layer: 1792 neural units; third hidden layer: 1920 neural units; fourth hidden layer: 1792 neural units; fifth hidden layer: 1536 neural units; output layer: n neural units, where the value of n is determined by the type of fault.

[0049] It is understandable that the creation of the training set adopts the back propagation algorithm, which is an optimization algorithm that enables the neural network to learn the mapping relationship between input and output by continuously adjusting the connection weights between each neuron in the network. Specifically, the back propagation algorithm calculates the state and activation value of each layer, advances the calculation error from the last layer forward, and updates the parameters to minimize the error between the predicted output and the actual output of the network. This process will continue to iterate until the stopping criterion is met.

[0050] It can be understood that in step S5, the fault interval judgment based on the classification result combined with the line topology diagram includes: when the fault interval is determined, the distance to the fault point is processed, which is mainly calculated based on some physical characteristics, wherein the traveling wave speed of the overhead line is about 290m / us, and the traveling wave speed of the power cable is about 170m / us; there are two calculation methods, one is the single-end traveling wave method: the fault is located by the time difference of the traveling wave going back and forth to the fault point; the other is the double-end traveling wave method: the fault is located by the time difference of the traveling wave reaching both sides of the line; In this step, it is necessary to accurately locate the wave head position of the traveling waves uploaded by two different devices. The slope judgment method is used here for calculation. The judgment method is divided into two steps: the first step is to determine the interval range of the drastic change of the traveling wave and intercept the 1000 points before and after it. This step will mainly use a low-pass filter to process it and obtain the rough range of the drastic fluctuation of the traveling wave; the second step, after determining the rough range of the fluctuation, start to calculate the wave head point, which is further as follows: 1. De-noise and curve fit the interval data; 2. Sampling is performed at the reporting frequency of the original data reporting point; 3. It calculates the slope value of the previous point and the next point of the sampling data, and judges it with the threshold. If it is greater than the threshold, it is a candidate point and is recorded in the candidate set; if it is less than the threshold, it is not processed; When all candidate sets in the two devices are determined, the earliest data with the smallest difference is obtained as the final time difference by comparing the data in the candidate sets, and then the relevant fault point location is calculated.

[0051] It can be understood that the step S1 specifically includes the following steps: Step 1.1, equipment installation. Before installation, you must fully understand the specific conditions of the lines to be installed, including but not limited to the following: 1. Line topology diagram. The topology diagram marks the line direction, length, geographical location, switch location and type, main line and branch topological relationship, conductor type, cross-sectional area, main line and branch design load size, and whether it is a mixed line; a detailed topology diagram is an important basis for the line configuration plan.

[0052] 2. Spacing table. The spacing table marks the front and rear spacing of the line towers. Not all units can provide it. Accurate spacing can reduce the distance measurement error. When the spacing table cannot be provided, the average spacing of 50m is used for calculation.

[0053] 3. Line fault operation status. Such as: the normal operating current of the main line and important branches (determines whether the terminal can be online for a long time), fault frequency, fault type, high-incidence areas of faults, etc.

[0054] Step 1.2: The terminal configuration principles are as follows: 1. At the exit of the substation line; 2. Where the cable line and the overhead line are connected, if there is a cable line in the middle, both sides should be installed; 3. If the main line of the overhead line exceeds 5km, one more set of equipment shall be added; 4. For T-connected lines, if the branch line is longer than 1.5 km, one set is usually installed on the tower at the head end of the branch line; 5. For lines with more users or higher load density, the number of installations can be appropriately increased; 6. Appropriately increase the number of configurations for important line loads; 7. Cross-regional demarcation points; 8. When installed on the same pole as the switch, it can be adjusted to the rear pole appropriately; Step 1.3: Determine the data upload protocol, device upload topic address, XOL / IMEI number / WAVEFORM_DATA and other data.

[0055] It can be understood that the step S2 specifically includes the following steps: Step 2.1, denoising is performed by an improved discrete wavelet transform algorithm, the improved DWT algorithm decomposes the signal into approximate coefficients and detail coefficients of different scales, that is, the improved DWT uses discrete wavelet functions and discrete time scales, and realizes signal decomposition through filtering and downsampling operations. The discrete data signal is decomposed into a group of high-frequency and low-frequency sub-signals. And the corresponding signal-to-noise ratio data is obtained.

[0056] like Figure 1 As shown, the hardware device is provided with a terminal module, and the terminal module is provided with a timing module, a processor, a 4G module, an electric field module, a traveling wave / current module and a lithium battery pack.

[0057] Step 2.2, use the improved PSO algorithm to perform secondary denoising, set the iterative termination condition, perform multiple iterative processes, and always compare it with the signal-to-noise ratio result of step 3.1. If the difference meets the termination condition, it means that the denoising process is completed. Otherwise, modify the PSO initial parameters and continue to iterate.

[0058] It can be understood that the step S3 specifically includes the following steps: step 3.1, using a polynomial fitting algorithm to perform an initial fitting on the result obtained in step 3, where an n-order polynomial is used, and the desired result is that all point sets are approximately connected; Step 3.2, using the cumulative curve fitting approximation (CCFA) algorithm to fit the result of step 3.1, and further perform denoising, and perform supplementary processing on the result of step S2.

[0059] It can be understood that the use of the traveling wave fault location algorithm to achieve accurate positioning of the fault point specifically includes the following steps: Determine the range of the interval where the traveling wave changes violently, and intercept 1000 points before and after it. This step will use a low-pass filter to process it and obtain the rough range of the traveling wave fluctuation; De-noising and curve fitting are performed on the interval data; Sampling is performed at the reporting frequency of the data's original reporting point; By calculating the slope value of the previous point and the next point of the sampling data, and judging it with the threshold, if it is greater than the threshold, it is a candidate point and recorded in the candidate set; if it is less than the threshold, no processing is done; When all candidate sets in the two devices are determined, the earliest data with the smallest difference is obtained as the final time difference by comparing the data in the candidate sets, and then the relevant fault point location is calculated. The calculation method is: Single-ended traveling wave method: Fault location is performed based on the time difference between the traveling wave and the fault point: X=1 / 2(tmf-tm)*V; Double-terminal traveling wave method: Fault location is performed based on the time difference between the traveling wave reaching both sides of the line: 2X=L+Vtm-Vtn; X=1 / 2+(tm-tn)V / 2; Then the final fault location is obtained.

[0060] The specific description of the present invention in the above embodiments is only used to further illustrate the present invention and cannot be understood as limiting the scope of protection of the present invention. Technical engineers in this field may make some non-essential improvements and adjustments to the present invention based on the contents of the above invention, which fall within the scope of protection of the present invention.

Claims

1. A method for identifying and locating distribution line faults based on DWT and BP algorithm, characterized in that: The following steps are involved: S1, install the hardware equipment on different phases of the nodes of the same line in the distribution network line, collect and record the discrete data information of current, voltage and traveling wave in the overhead line, and upload it to the platform database through the MQTT protocol; S2, build the network topology of the line to be monitored according to the hardware device mounting points, and use the particle swarm optimization algorithm and DWT algorithm to pre-process the original data for noise reduction; S3, performing polynomial fitting and cumulative curve fitting approximation processing on the data after noise reduction preprocessing to obtain continuous curve data after fitting; S4, performing multiple truncation processing on the fitted continuous curve data, and analyzing the frequency components in the truncation interval by fast Fourier transform, and extracting the peak information in the frequency domain as the period information; S5, build a BP neural network model, input the preprocessed data into the network for training and classification recognition, judge the fault interval based on the classification results combined with the line topology map, and use the traveling wave fault ranging algorithm to locate the fault point.

2. A method for identifying and locating distribution line faults based on DWT and BP algorithm according to claim 1, characterized in that: In step S2, the combination of the particle swarm optimization algorithm and the DWT algorithm for denoising the original data includes: denoising by the DWT algorithm, the DWT algorithm decomposing the signal into approximate coefficients and detail coefficients of different scales, the DWT algorithm using discrete wavelet functions and discrete time scales, and realizing signal decomposition by filtering and downsampling operations, decomposing the original signal into a group of high-frequency sub-signals and low-frequency sub-signals; improving the Heaviside projection function in the DWT algorithm, and optimizing the basic Heaviside function using the sparse regularized Heaviside function; It also includes adding a signal-to-noise ratio comparison module based on the PSO algorithm, and iterating the results processed by the improved DWT algorithm again.

3. A method for identifying and locating distribution line faults based on DWT and BP algorithm according to claim 2, characterized in that: In step S2, it also includes: performing sparse regularization optimization on the Heaviside projection function in the DWT algorithm to improve the denoising effect on nonlinear and non-smooth data.

4. A method for identifying and locating distribution line faults based on DWT and BP algorithm according to claim 1, characterized in that: In step S3, the cumulative curve fitting approximation processing includes: using a polynomial fitting algorithm to perform preliminary fitting on discrete data, performing secondary fitting through a cumulative curve fitting approximation algorithm, comparing fNIRS data features with similar data features collected by the hardware device, using the behavior of the approximate signal curve in the sliding window, obtaining an estimated curve from multiple overlapping windows, and using a weighted logical recombination method to receive the filtered signal to obtain the final fitting curve.

5. A method for identifying and locating distribution line faults based on DWT and BP algorithm according to claim 1, 2, 3 or 4, characterized in that: Between step S4 and step S5, a processing step is also included: performing an iterative operation on the operation in S4, and when the peak information in the frequency domain after the fast Fourier transform tends to be consistent, terminating the iterative operation to obtain a final peak information set.

6. A method for identifying and locating distribution line faults based on DWT and BP algorithm according to claim 5, characterized in that: In step S4, the multiple truncation processing of the fitted continuous curve data includes: preferentially setting a preliminary truncation threshold, performing fast Fourier transform on the truncated data, analyzing the frequency components of the data in the truncation interval, obtaining the peak information in the frequency domain, performing statistical processing on all the peak information, obtaining a preliminary peak information set, and iterating the truncation threshold, performing multiple iterations, and obtaining a new peak information set; after multiple iterations, if the peak information obtained after multiple iterations is maintained within a certain range and the frequency of occurrence is greater than a certain threshold, the peak information is defined as periodic information.

7. A method for identifying and locating distribution line faults based on DWT and BP algorithm according to claim 6, characterized in that: According to the operation in the processing step, the continuous curve data of n cycles are intercepted from the original data in step S3, and sampling processing is performed to obtain discrete waveform data.

8. A method for identifying and locating distribution line faults based on DWT and BP algorithm according to claim 7, characterized in that: In step S5, the building of the BP neural network model includes: creating a training set and a test set, inputting them into the BP neural network for iterative training, obtaining the model parameters of the BP neural network and saving them.

9. A method for identifying and locating distribution line faults based on DWT and BP algorithm according to claim 8, characterized in that: In step S5, it also includes: performing actual test verification on the BP neural network model, and optimizing the neuron parameters, loss function, optimizer, initialization weights and network structure according to the test verification results to obtain the final waveform classification model.

10. A method for identifying and locating distribution line faults based on DWT and BP algorithm according to claim 9, characterized in that: In step S5, it also includes: when the classification result of the final waveform classification model is a fault defect, defect interval positioning is performed: the defect interval positioning includes: a short circuit fault interval judgment algorithm and a single-phase grounding fault interval judgment algorithm.

Images