Power optical cable fault detection and positioning method based on extreme learning machine

Abstract

The invention relates to the field of power optical cable facilities, in particular to a power optical cable fault detecting and positioning method based on an extreme learning machine. The OTDR signal is subjected to multi-scale decomposition and de-noising processing based on wavelet packet analysis, and high-frequency noise is effectively suppressed and key event point features are reserved by refining frequency band division, so that the signal quality is improved. And constructing a fault classification model based on an extreme learning machine, and realizing efficient classification and fault mode identification of de-noised OTDR data. And in combination with a white whale optimization algorithm (BWO), carrying out optimization search on a fault detection result output by the classification model so as to improve the detection precision and algorithm robustness in a complex optical cable environment. And an improved Dijkstra algorithm is introduced to realize accurate positioning and path tracking, so that the accuracy and response speed of fault positioning are remarkably improved. Through multi-level algorithm fusion, the problems of signal noise interference, insufficient pattern recognition accuracy and low positioning precision in a complex network environment in optical cable fault detection are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of power optical cable facilities, namely, a method for power optical cable fault detection and location based on extreme learning machine. Background Art

[0002] With the development of the intellectualization and informatization of the power system, the importance of optical cables in the power communication system has become increasingly prominent. The safe and stable operation of optical cables is directly related to the reliability and efficiency of the power grid. Therefore, the rapid detection and accurate location of power optical cable faults are particularly important, which can not only shorten the power outage time, improve the continuity of power supply, but also effectively reduce the maintenance cost. However, the traditional fault detection methods based on manual inspection and experience judgment can no longer meet the high-efficiency and accurate requirements of modern power systems.

[0003] In the aspect of optical cable fault detection and location, many advanced technologies have been proposed and applied. Among them, the fault detection based on OTDR obtains the event point information in the optical cable by sending optical pulses and analyzing the returned signals. However, the traditional method is affected by noise and interference, and the detection accuracy is limited. The fault classification based on machine learning uses historical data to train the model to identify new faults, but there are challenges in the face of large-scale data and high-dimensional features. The optical cable fault location based on GIS can convert the distance detected by OTDR into the actual geographical location to achieve accurate location, but the existing methods have low automation. Although these methods have made certain progress, there is still room for improvement in improving the detection accuracy and efficiency in complex environments.

[0004] Due to the environmental complexity and external interference, traditional OTDR data analysis often faces the problem of increasing noise level, resulting in a decrease in the accuracy of fault point identification, which in turn affects the reliability and stability of the power system. When the existing optical cable fault mode recognition algorithms process OTDR data, the event point classification is prone to be inaccurate. The traditional manual analysis method is restricted by subjective factors and has a slow processing speed, which not only increases the complexity of fault diagnosis, but also prolongs the troubleshooting time and affects the operation and maintenance efficiency. In addition, in the geographical information analysis of fault location, due to inaccurate location matching and the limitation of algorithm processing ability, the relative position error is often large, and it is difficult for the emergency repair team to reach the fault point in time, which prolongs the recovery time. The existing fault location and diagnosis highly rely on manual on-site testing, resulting in cumbersome operation, low efficiency, increased labor cost and response time, making it difficult for modern power systems to achieve efficient fault detection, accurate location and rapid response.

[0005] Although these methods have their respective optimizations, they still face limitations such as poor noise reduction, inaccurate classification, insufficient optimization accuracy and slow training speed, and there is an urgent need to improve in terms of signal-to-noise ratio improvement, event point location and fault classification. Summary of the Invention

[0006] The object of the present invention is to propose a power optical cable fault detection and positioning method based on an extreme learning machine, which has good signal noise reduction effect, accurate fault classification, and high fault point positioning accuracy in optical cable fault detection, covering multi-level analysis of OTDR data and fault positioning strategies.

[0007] The technical solution of the present invention is: a power optical cable fault detection and positioning method based on an extreme learning machine. First, perform multi-scale decomposition and denoising processing on the OTDR signal based on wavelet packet analysis. By refining the frequency band division, effectively suppress high-frequency noise and retain the characteristics of key event points, thereby improving the signal quality. Then, construct a fault classification model based on the extreme learning machine (ELM), and use its powerful non-linear mapping ability and fast training characteristics to achieve efficient classification of the denoised OTDR data and fault mode recognition. Subsequently, combined with the beluga whale optimization algorithm (BWO), optimize the search for the fault detection results output by the classification model to improve the detection accuracy and algorithm robustness in a complex optical cable environment. Finally, introduce an improved Dijkstra algorithm, optimize the path search efficiency by combining a binary sorting tree structure, quickly lock the fault position in the optical cable, and achieve accurate positioning and path tracking, thereby significantly improving the accuracy and response speed of fault positioning.

[0008] In the above solution, it further includes:

[0009] The OTDR noise data processing method based on wavelet packet includes decomposing the original OTDR data by applying wavelet transform; then, performing threshold processing on the wavelet coefficients to achieve signal-noise separation; finally, reconstructing the wavelet packet coefficients of each frequency band to obtain the denoised OTDR signal.

[0010] The fault detection model based on the beluga whale optimization algorithm is to optimize the input weights and bias parameters in the ELM by the common beluga whale optimization algorithm (BWO), which can significantly reduce the error of OTDR measurement data classification, and improve the robustness and accuracy of classification, thereby improving the fault classification effect of OTDR data.

[0011] The optical cable fault positioning method based on Dijkstra includes the following steps:

[0012] (1) Rough positioning: Use the improved Dijkstra algorithm, combined with the binary sorting tree data structure, to optimize the path search efficiency in a complex network, determine the shortest path from the fault detection point to the potential fault point, and initially determine the approximate location of the fault.

[0013] (2) Model prediction: Input the relevant data of the preliminarily determined fault area into the pre-trained Light GBM model, and use the mapping model trained with historical fault data to predict a more accurate fault location.

[0014] (3) Fine search: Near the location predicted by the Light GBM model, apply the improved bacterial foraging algorithm (IBFA) for local optimization search to further improve the positioning accuracy until the exact fault point is found.

[0015] The beneficial effects of the present invention are as follows: 1. Through the multi-level algorithm fusion, the present invention solves the problems of signal noise interference, insufficient pattern recognition accuracy, and low positioning accuracy in complex network environments in optical cable fault detection. In the detection stage, a multi-scale denoising method based on wavelet packets is adopted to effectively suppress high-frequency noise, retain the characteristics of key event points, and combine with the ELM model to quickly identify different fault modes, improving the accuracy and robustness of detection. In addition, the BWO is used to further optimize the fault detection results, enhancing the search efficiency and recognition effect of the algorithm in non-linear environments. 2. In the positioning stage, the present invention innovatively introduces an improved Dijkstra algorithm and combines it with a binary sorting tree structure, significantly improving the path search efficiency and fault positioning accuracy. The algorithm can quickly lock the fault location according to the change characteristics of the OTDR signal, solving the problems of insufficient accuracy and large computational overhead of traditional positioning methods in complex optical cable network environments. The overall solution significantly improves the intelligence and efficiency of the optical cable fault diagnosis system through the full-process optimization from data denoising, fault detection to precise positioning, providing reliable technical support for optical cable maintenance and operation and maintenance decision-making. 3. Through multi-level intelligent diagnosis and optimization strategies, the full-process efficiency and accuracy of power optical cable fault detection and positioning are improved. 4. Compared with the prior art, the present invention has been comprehensively optimized in terms of signal processing, fault classification, optimization search, and positioning efficiency, significantly improving the algorithm robustness and response speed, providing an efficient, accurate, and robust comprehensive solution for power optical cable fault detection and positioning, effectively enhancing the reliability and operation efficiency of the power system, and at the same time reducing the maintenance cost.

[0016] The following will further describe the embodiments of the present invention in detail with reference to the accompanying drawings. Description of the Drawings

[0017] Figure 1 is the flow chart of the present invention.

[0018] Figure 2 is the flow chart of fault detection and positioning.

[0019] Figure 3 is the flow chart of the Dijkstra algorithm.

[0020] Figure 4It is the flowchart of the IBFA algorithm. Specific implementation mode

[0021] See Figures 1-4 , a power optical cable fault detection and location method based on extreme learning machine, including OTDR noise data processing based on wavelet packet, OTDR classification method model based on extreme learning machine, fault detection model based on beluga optimization algorithm, and optical cable fault location method based on Dijkstra.

[0022] 1. OTDR noise data processing based on wavelet packet

[0023] The present invention proposes a denoising method based on wavelet transform and wavelet packet analysis to improve the quality of OTDR signals and the accuracy of fault detection. In OTDR testing, factors such as the measurement environment and instrument accuracy will introduce noise, which not only masks the characteristics of key event points (such as fiber joints, fault points, and fiber ends), but also affects the identification and location of fault modes. Therefore, it is necessary to denoise the OTDR data to improve the readability and accuracy of the signals.

[0024] First, apply wavelet transform to decompose the original OTDR data.

[0025] The OTDR signal contains multiple event points, and each event point corresponds to a specific position in the optical fiber. However, the noise components are usually concentrated in the high-frequency part and will cover the characteristics of the event points. Wavelet transform can decompose the signal into wavelet coefficients containing different frequencies and time scales, where the low-frequency coefficients mainly reflect the trend changes and event point information of the signal, and the high-frequency coefficients mainly contain noise components. Let w be the wavelet coefficient, and a set of wavelet coefficients is obtained through decomposition, and these coefficients contain the characteristics of the signal in different frequency bands.

[0026] Next, perform threshold processing on the wavelet coefficients to separate the signal from the noise.

[0027] The selection of the threshold T directly affects the denoising effect. In OTDR data, if the selected threshold is too high, useful signal components may be cut; if the threshold is too low, too much noise will be retained. Common threshold processing methods include soft threshold and hard threshold. The soft threshold processing uses the following formula:

[0028]

[0029] Among them, is the processed coefficient, w is the original wavelet coefficient, T is the threshold, and sign(w) is the sign function, which is used to retain the positive and negative of the wavelet coefficient. The soft threshold method can smooth the signal, reduce the interference of high-frequency noise on the OTDR signal, and retain the main event point characteristics. The formula for hard threshold processing is as follows:

[0030]

[0031] Among them, is the processed coefficient, w is the original wavelet coefficient, and T is the threshold. The hard threshold method directly sets the coefficients less than the threshold to zero and keeps the coefficients greater than the threshold unchanged. This method performs well in preserving the sharp features (such as mutation points and reflection peaks) of the OTDR signal and is suitable for scenarios of detecting fiber optic connectors and fault points.

[0032] After threshold processing, the wavelet inverse transform is applied to reconstruct the denoised OTDR signal. However, the OTDR signal usually contains multiple complex event points, and the noise is unevenly distributed in different frequency bands. It is difficult to completely remove the noise only using wavelet transform. Therefore, wavelet packet analysis is introduced to further improve the denoising effect. Wavelet packet analysis performs multi-scale decomposition on the signal, not only decomposing the low-frequency part but also performing multi-layer decomposition on the high-frequency part, thereby capturing more subtle changes in the OTDR signal. The wavelet packet decomposition satisfies the principle of energy conservation:

[0033]

[0034] Among them, E signal is the original OTDR signal, that is, the intensity of the echo signal in the optical fiber changes with time; x[i] is the discrete sampling value of the original signal, where i represents the index of the sampling point. N is the total number of sampling points of the signal. It represents the length of the signal; J is the total number of layers of wavelet packet decomposition, and each layer contains wavelet coefficients of different frequency components;

[0035] W j,k is the wavelet coefficient at the j-th layer and the k-th position. It is the representation of the signal in a specific frequency band and scale. This formula shows that the total energy of the coefficients of each decomposed layer is equal to the total energy of the original signal, ensuring the fidelity of the signal and the reliability of the denoising effect.

[0036] Finally, the wavelet packet coefficients of each frequency band are reconstructed to obtain the denoised OTDR signal. To quantify the denoising effect, the energy characteristics of each frequency band are calculated and normalized. The normalized energy feature vector can be used to detect event points and reflection peaks in the OTDR signal, providing a reliable basis for fault location.

[0037] The denoising method proposed by the present invention combines wavelet transform, threshold processing, and wavelet packet analysis, which can effectively suppress the high-frequency noise in the OTDR signal and improve the readability of the signal and the accuracy of event point recognition. Experimental results show that this method has good robustness in a complex noise environment, providing reliable technical support for optical cable fault detection and maintenance.

[0038] 2. OTDR Classification Method Model Based on Extreme Learning Machine

[0039] In the optical fiber communication network of the power system, OTDR is a key tool for detecting fiber optic link faults and locating fault points. OTDR analyzes the link status by transmitting pulse signals into the optical fiber and detecting the reflected signals. However, due to the complexity of the actual optical fiber link, the measurement signals are usually interfered by various noises, and it is difficult for traditional analysis methods to effectively distinguish between normal and fault states. Therefore, extreme learning machine (ELM) is used to classify OTDR signals to improve the accuracy and efficiency of fault detection.

[0040] ELM is a single-hidden layer feedforward neural network (SLFN), which consists of an input layer, a hidden layer, and an output layer. The input weights and hidden layer biases of ELM are both randomly generated, and the output layer weights are calculated by the generalized inverse matrix, so as to quickly obtain the prediction results. Let the training data set contain N OTDR measurement samples, denoted as (x i , t i ). Among them, x i = [x i1 , x i2 , …, x in T ∈ R n represents the feature vector of the i-th sample. This feature vector contains key features extracted from OTDR measurement data, such as reflection intensity and offset time, etc. t i = [t i1 , t i2 , …, t im T ∈ R m is the corresponding target output, representing the fault type, such as fiber breakage, connection loss, and bending loss, etc. For an SLFN with L hidden layer nodes, the output can be expressed as:

[0041]

[0042] where b i is the bias of the i-th hidden layer node; w i · x j is the inner product of the input weight w i and the input feature x j ; g(x) is the activation function; β i is the weight of the output layer; o j is the network output of the j-th sample, used to predict its fault type.

[0043] The training objective of ELM is to minimize the output error, so that the predicted output o j is close to the target output t j ​​As consistent as possible. This goal can be expressed as:

[0044]

[0045] This means that there is a set of w i , β i and b i such that:

[0046]

[0047] It can be represented in matrix form as:

[0048] Hβ = T (2-4)

[0049] where H is the output matrix of the hidden layer, containing the results of the non-linear mapping of all samples through the hidden layer; is the output weight vector; T is the target output matrix, representing the true fault types of the samples.

[0050] Specifically, H can be expanded as:

[0051]

[0052] During the training process, the goal of ELM is to determine the output weights by solving the linear system Hβ = T. Its optimal solution is:

[0053]

[0054] where H + is the generalized inverse matrix of H, used to calculate the least squares solution.

[0055] In ELM, the activation function g(x) is usually selected as a non-linear function. In this invention, the Sigmoid function is adopted, and its expression is:

[0056]

[0057] However, since the input weights w i and the bias b i in ELM are randomly generated, it may cause the model to not guarantee the global optimal solution and is prone to falling into the local optimum, thus affecting the fault classification performance. Therefore, this invention introduces BWO to optimize the input weights w i and the bias b i to improve the accuracy and robustness of the model in OTDR data classification.

[0058] 3. Fault Detection Model Based on Beluga Whale Optimization Algorithm

[0059] OTDR measurement data usually contains a large amount of noise and exhibits highly non - linear characteristics. Traditional classification methods are difficult to achieve high accuracy in complex environments. To improve the classification performance of ELM, the Beluga Whale Optimization algorithm (BWO) is used to optimize the input weights and bias parameters in ELM, thus improving the fault classification effect of OTDR data.

[0060] BWO is a population - based meta - heuristic algorithm inspired by the swimming, hunting, and falling behaviors of beluga whales. BWO consists of three stages: exploration, exploitation, and falling, combined with an adaptive balance factor and Lévy flight strategy to enhance the global search ability and local exploitation ability. In BWO, each beluga whale individual is regarded as a candidate solution in the parameter space of the ELM model. During the optimization process, the positions of the beluga whale population are updated iteratively to minimize the classification error of ELM.

[0061] The beluga whale population can be represented as a matrix X:

[0062]

[0063] where n is the population size, that is, the number of beluga whales, and d is the dimension of the ELM model parameters, including input weights and biases. Each individual x i represents a set of candidate parameter configurations of the ELM model, and the fitness value F X is used to evaluate the performance of each individual in the OTDR fault classification task:

[0064]

[0065] where f(·) is the classification error function of the ELM model, representing the error value in the OTDR data classification task. The balance factor B f is used to control the switching between the exploration and exploitation stages, and its expression is:

[0066]

[0067] B f is the balance factor, which determines whether the algorithm is in the exploration stage or the exploitation stage; T is the current iteration number, T max is the maximum iteration number, B0∈(0,1) is the initial value. If B f > 0.5, the algorithm enters the exploration stage; if B f ≤0.5, it enters the exploitation stage. As the number of iterations increases, the fluctuation range of B f shrinks from (0,1) to (0,0.5), gradually enhancing the exploitation ability. In the exploration stage, the position update of the beluga whale is through simulating paired swimming behavior, and the update formula is as follows:

[0068]

[0069] Among them, is the new position of the i-th individual in the p j dimension; r1 and r2 are random numbers, simulating the random swimming behavior of beluga whales; and are the positions of the current i-th and r-th individuals.

[0070] In the development stage, to enhance search diversity, BWO introduces the Lévy flight strategy, and the update formula is:

[0071]

[0072] Among them, is the new position of the i-th individual; is the best position in the current population; C1 is the jump intensity, defined as:

[0073]

[0074] L F is the Lévy flight function, and its expression is:

[0075]

[0076] Among them, μ and v are normally distributed random numbers; β = 1.5 is the parameter of the Lévy distribution, and σ is the standard deviation of the Lévy distribution.

[0077] In the falling stage, to avoid the algorithm falling into local optimum, BWO simulates the falling behavior of beluga whales, and the update formula is:

[0078]

[0079] X step is the falling step size, defined as:

[0080]

[0081] Among them, u b and l b are the upper and lower limits of the parameters; C s is the step size factor, and its expression is:

[0082] C s = 2W f ×n(3 - 10)

[0083] W f is the falling probability, and the calculation formula is:

[0084]

[0085] Optimizing the input weights and biases of ELM through BWO can significantly reduce the error of OTDR measurement data classification and improve the robustness and accuracy of classification. The optimized BWO-ELM model can better adapt to the complex non-linear environment in the optical fiber link, significantly improving the accuracy and efficiency of fault detection in the optical fiber communication link of the power system.

[0086] 4. Dijkstra-based Optical Cable Fault Location Method

[0087] The specific steps are as follows:

[0088] (1) Rough Location: Using the improved Dijkstra algorithm and combining with the binary sorting tree data structure to optimize the path search efficiency in complex networks, determine the shortest path from the fault detection point to the potential fault point, and initially determine the approximate location of the fault.

[0089] (2) Model Prediction: Input the relevant data of the initially determined fault area into the pre-trained Light GBM model, and use the mapping model trained with historical fault data to predict a more accurate fault location.

[0090] (3) Fine Search: Near the location predicted by the Light GBM model, apply the improved bacterial foraging algorithm (IBFA) for local optimization search to further improve the location accuracy until the exact fault point is found.

[0091] Through the combination and cooperation of the above steps (see Figure 2 ), it can not only significantly improve the accuracy of fault detection, ensure that the fault area is quickly and accurately identified, but also greatly improve the efficiency and accuracy of fault location, providing strong technical support for the rapid repair of the optical cable network, enabling maintenance personnel to quickly reach the fault site, reducing the impact of the fault on the communication network, and thus ensuring the stability and reliability of communication services.

[0092] In the optical cable network, in order to achieve precise location of the fault point, an integration strategy of multiple advanced algorithms is adopted. First, based on the improved Dijkstra algorithm, using the binary sorting tree data structure to optimize its efficiency in complex networks, it can quickly determine the shortest path from the fault detection point to the actual fault point, thus quickly locking the fault area in the huge network topology. The steps of the improved Dijkstra algorithm are as Figure 3 shown:

[0093] First, in order to quickly lock the potential fault area from the fault detection point in the optical cable network, an improved Dijkstra algorithm is used, and the path search process is optimized by combining with the binary sorting tree structure. When initializing the algorithm, an open vertex set Open and a closed set Close are constructed. In the initial state, Close only contains the starting point S. By obtaining the adjacent nodes of the starting point and sorting the adjacent nodes according to the loss weight of the OTDR echo signal. The binary sorting tree structure is used to select the intermediate node L with the minimum loss, and move it from the Open set to the Close set, while updating the weight information of the adjacent nodes. Repeat this process until the Open set is empty. Finally, according to the recorded parent nodes, the shortest path from the detection point to the fault point is traced to achieve preliminary positioning.

[0094] After obtaining the preliminary fault area, in order to further improve the positioning accuracy, the LightGBM model is used to predict the fault location. LightGBM is an efficient gradient boosting decision tree model that can handle complex non-linear mapping relationships. During the training process, a large amount of historical OTDR echo signal data is collected as feature inputs, including the amplitude of the reflection peak, the position of the event point, and the loss attenuation rate, etc. By marking the fault location, the mapping relationship between the OTDR signal features and the actual fault location is established. After inputting the new OTDR test data, the LightGBM model uses the trained mapping relationship to make predictions and outputs the fault location, providing support for further fine search. Near the location predicted by LightGBM, in order to improve the positioning accuracy and find the exact fault point, the present invention introduces IBFA for local optimization search. The process is as follows, see Figure 4 。

[0095] IBFA simulates the foraging behavior of bacteria in nature and can perform global optimization search in the solution space of multi-dimensional non-linear problems. First, initialize the position vector Z of the bacteria population = {z1, z2,..., z m}, where z m represents the position of the m-th bacteria. By calculating the brightness φ0 of each bacteria as the evaluation function value of its initial state, where the brightness attenuation is determined by the light intensity absorption coefficient γ and the distance r between bacteria, and the formula is expressed as The velocity vector μ α,β (ζ) of the bacteria is updated in each iteration, and the formula is as follows:

[0096]

[0097] Among them, ψ is the inertia weight that controls the speed update; β0 is the initial attraction of the bacteria; γ ζ is the light intensity absorption coefficient of the ζ-th generation; l β (ζ) and l α($\zeta$) represents the brightness positions of the $\beta$-th and $\alpha$-th bacteria in the $\zeta$-th generation; $\alpha_0rand(-, +)$ is a random perturbation term used to increase search diversity. To further improve the optimization effect, IBFA performs a mutation operation on the optimal individual, and the formula is as follows:

[0098]

[0099] Among them, $Q$ T is the random mutation probability, which controls the occurrence of mutation. IBFA continuously iterates and searches until the preset maximum number of iterations is reached or the convergence condition is satisfied, and finally outputs the optimal fault point position. Through the comprehensive application of the improved Dijkstra algorithm, the LightGBM model, and IBFA, the method of the present invention has shown significant performance improvement in the optical cable OTDR fault location task. First, the Dijkstra algorithm efficiently narrows the fault search range; then, the LightGBM model uses historical data for rapid prediction; finally, IBFA further precisely searches for the fault point position within the local area.

[0100] The above description is only the specific implementation manner of the present invention, and various examples do not constitute a limitation to the essence of the present invention.

Claims

1. A method for detecting and locating power optical cable faults based on extreme learning machine, characterized in that Firstly, the OTDR signal is decomposed and denoised at multiple scales based on wavelet packet analysis. By refining the frequency band division, high-frequency noise is effectively suppressed and the key event point features are retained, thereby improving the signal quality. Then, a fault classification model based on extreme learning machine is constructed. By utilizing its powerful nonlinear mapping ability and fast training characteristics, efficient classification and fault mode recognition of denoised OTDR data are achieved. Subsequently, the fault detection results output by the classification model are optimized and searched in combination with the Beluga optimization algorithm to improve the detection accuracy and algorithm robustness in complex optical cable environments. Finally, the improved Dijkstra algorithm is introduced to optimize the path search efficiency by combining the binary sorting tree structure, quickly lock the fault location in the optical cable, and achieve precise positioning and path tracking, thereby significantly improving the accuracy and response speed of fault location.

2. A method for detecting and locating power optical cable faults based on an extreme learning machine according to claim 1, characterized in that The wavelet packet-based OTDR noise data processing method comprises applying wavelet transform to decompose the original OTDR data; then, threshold processing is performed on the wavelet coefficients to achieve signal and noise separation; finally, the wavelet packet coefficients of each frequency band are reconstructed to obtain the denoised OTDR signal.

3. A method for detecting and locating power optical cable faults based on an extreme learning machine according to claim 1, characterized in that The fault detection model based on the Beluga optimization algorithm is a commonly used Beluga optimization algorithm to optimize the input weights and bias parameters in the ELM, which can significantly reduce the error of OTDR measurement data classification and improve the robustness and accuracy of classification, thereby improving the fault classification effect of OTDR data.

4. A method for detecting and locating power optical cable faults based on an extreme learning machine according to claim 1, characterized in that The optical cable fault location method based on Dijkstra comprises the following steps: (1) Rough positioning: Using the improved Dijkstra algorithm combined with the binary sorting tree data structure, the path search efficiency in complex networks is optimized, the shortest path from the fault detection point to the potential fault point is determined, and the approximate location of the fault is preliminarily determined; (2) Model prediction: The relevant data of the initially determined fault area is input into the pre-trained Light GBM model, and a more accurate fault location is predicted using the mapping model trained with historical fault data; (3) Fine search: Near the location predicted by the Light GBM model, the improved bacterial foraging algorithm (IBFA) is applied to perform local optimization search to further improve the positioning accuracy until the exact fault point is found.

Images