Rural power grid power distribution intelligent monitoring and fault positioning system based on edge calculation

Abstract

The invention relates to a rural power grid power distribution intelligent monitoring and fault positioning system based on edge calculation. The system comprises a data acquisition and preprocessing module used for acquiring time sequence data of each edge node of the power distribution network based on a unified protocol and performing standardization processing on the time sequence data to obtain standard time sequence data; the fault feature analysis module is used for analyzing abnormal features of the voltage and the current based on a preset fault detection model according to the standard time sequence data to obtain fault feature data; the preliminary fault positioning module is used for calculating probability distribution of a fault position based on a preset fault positioning model according to the fault feature data and the topology identifier to obtain preliminary fault positioning data; and the fault positioning correction module is used for correcting a fault positioning result according to the initial fault positioning data and the operation environment parameters to obtain a fault positioning result. By adopting the system, the detection capability and positioning precision of early weak fault features can be improved, and the system adapts to rural power grid complex environments and resource limited conditions.

Description

Technical Field

[0001] This invention belongs to the field of power system monitoring and fault diagnosis technology, and in particular relates to a smart monitoring and fault location system for rural power grid distribution based on edge computing. Background Technology

[0002] With the development of smart grid and Internet of Things technologies, intelligent and refined operation and maintenance of distribution networks has become an important trend. In rural distribution networks (rural power grids), due to the wide distribution of lines, complex operating environment, and large load fluctuations, achieving efficient and accurate fault monitoring and rapid fault location is crucial to ensuring power supply reliability.

[0003] In traditional technologies, rural power grid distribution monitoring mainly relies on centralized monitoring and data acquisition systems deployed in substations or control centers. Fault detection often employs overcurrent and undervoltage protection based on fixed thresholds, or depends on manual inspections by maintenance personnel and user reports. Fault location is typically achieved by analyzing the action sequence of protection devices and combining this with line topology to infer the fault location.

[0004] However, current centralized processing methods have significant drawbacks compared to conventional approaches: First, the massive amounts of monitoring data transmitted back to the central node for processing result in high communication bandwidth pressure and decision-making delays, making it difficult to meet the needs for rapid fault isolation and recovery. Second, simple threshold discrimination is susceptible to interference from load fluctuations and intermittent distributed power supply access, leading to a high false alarm rate and difficulty in detecting early, subtle fault characteristics. Furthermore, traditional location methods are not adaptable to changes in network topology and often fail to fully consider the impact of environmental factors such as temperature and humidity, and geographic spatial relationships on fault occurrence and propagation, resulting in limited location accuracy. In addition, rural power grid edge nodes have limited computing resources, making it difficult to directly deploy traditional complex algorithms. Therefore, there is an urgent need for an intelligent monitoring and fault location system that can process data locally, reduce response latency, integrate multi-source information, and improve location accuracy. Summary of the Invention

[0005] Therefore, it is necessary to provide a smart monitoring and fault location system for rural power distribution based on edge computing to address the above-mentioned technical problems.

[0006] In a first aspect, this application provides a smart monitoring and fault location system for rural power distribution based on edge computing, comprising:

[0007] The data acquisition and preprocessing module 11 is used to acquire time-series data of each edge node of the power distribution network based on a unified protocol, and to perform data cleaning, normalization and time alignment on the time-series data to obtain standard time-series data; wherein, the time-series data includes voltage, current, topology identifier and operating environment parameters;

[0008] The fault feature analysis module 12 is used to analyze the abnormal characteristics of voltage and current based on standard time series data and a preset fault detection model to obtain fault feature data; wherein, the fault feature data is used to characterize the probability of each edge node failing.

[0009] The preliminary fault location module 13 is used to calculate the probability distribution of the fault location based on the fault feature data and topology identifier and a preset fault location model to obtain preliminary fault location data; wherein, the preliminary fault location data is used to characterize the distribution of possible fault locations.

[0010] The fault location correction module 14 is used to correct the fault location result based on the preliminary fault location data and operating environment parameters to obtain the fault location result; wherein, the fault location result is used to indicate the location area of ​​the fault.

[0011] In one embodiment, the fault feature analysis module 12 includes:

[0012] The feature extraction subunit 121 is used to extract features from voltage and current sequences in standard time series data based on mathematical morphology gradient operators, capture transient change information of waveforms, and obtain gradient feature vectors.

[0013] The energy entropy calculation subunit 122 is used to calculate the energy entropy of the gradient feature vector within a preset time window to obtain the entropy feature value; wherein, the energy entropy is used to quantify the complexity and abrupt change intensity of the signal;

[0014] The initial fault score subunit 123 is used to input the gradient feature vector and entropy feature value into the preset lightweight random forest model for classification reasoning and output the initial fault score of each node.

[0015] The fault feature subunit 124 is used to normalize the initial fault score using the Sigmoid activation function to obtain fault feature data.

[0016] In one embodiment, the preliminary fault location module 13 includes:

[0017] The adjacency matrix construction sub-unit 131 is used to construct the adjacency matrix of the power distribution network based on the topology identifier;

[0018] Fault state vector construction subunit 132 is used to construct the fault state vector of each node based on fault characteristic data;

[0019] The fault propagation matrix calculation subunit 133 is used to calculate the fault propagation matrix based on the adjacency matrix and the fault state vector; wherein, the fault propagation matrix is ​​used to characterize the propagation intensity of fault features in the topological network; the expression of the fault propagation matrix is:

[0020]

[0021] in, This is the fault propagation matrix. It is an adjacency matrix. The Hadamard product represents the matrix multiplication, used for element-wise multiplication. For the fault state vector, It is a vector of all 1s. express transpose, This represents the absolute value operation. To prevent extremely small positive numbers with a denominator of zero;

[0022] The preliminary fault location subunit 134 is used to perform weighted summation and normalization on the row vectors of the fault propagation matrix to generate the fault probability distribution of each line section and obtain preliminary fault location data.

[0023] In one embodiment, the fault location and correction module 14 includes:

[0024] The environmental risk coefficient subunit 141 is used to extract the ambient temperature and ambient humidity from the operating environment parameters, and query the environmental risk coefficient corresponding to the current environmental conditions according to the preset historical fault statistics table.

[0025] Correction factor construction subunit 142 is used to construct a correction factor matrix based on environmental risk coefficients;

[0026] The fault probability correction subunit 143 is used to perform a dot product operation on the probability distribution and correction factor matrix in the preliminary fault location data to obtain the weighted corrected fault probability.

[0027] The fault location determination subunit 144 is used to select the area with the largest value in the weighted and corrected fault probability as the final fault point and output the fault location result.

[0028] In one embodiment, the fault preliminary location subunit 134 includes:

[0029] The normalization subunit 1341 is used to normalize each row vector of the fault propagation matrix using the L1 norm to obtain the node fault probability value of each node.

[0030] The line segment fault probability subunit 1342 is used to calculate the fault probability of each line segment based on the topological connection relationship of the power distribution network, and obtain the initial probability of the line segment.

[0031] The candidate fault screening subunit 1343 is used to filter the initial probability of the line segment through a preset dynamic truncation threshold, remove probability segments below the preset threshold, and obtain a set of candidate faults.

[0032] The preliminary fault location subunit 1344 is used to aggregate the segments in the candidate fault set using a density-based spatial clustering algorithm, and output the cluster with the highest probability density as the preliminary fault location data.

[0033] In one embodiment, the correction factor construction subunit 142 includes:

[0034] Geographic coordinate matrix construction sub-unit 1421 is used to obtain the geographic coordinates of each edge node in the power distribution network and construct the geographic coordinate matrix;

[0035] The spatial distance matrix calculation subunit 1422 is used to calculate the spatial distance between nodes using the Euclidean distance formula on the geographic coordinate matrix to obtain the spatial distance matrix.

[0036] The comprehensive correction factor calculation subunit 1423 is used to calculate the Gaussian radial basis function value as spatial weight based on the spatial distance matrix, and combine it with the environmental risk coefficient to generate the node comprehensive correction factor;

[0037] The expression for the comprehensive correction factor is as follows:

[0038]

[0039] In the formula, Represents a node The comprehensive correction factor, Represents a node Environmental risk coefficient and Representing nodes respectively and nodes The geographic coordinate vector, Represents a node The set of neighboring nodes, The bandwidth parameter represents the Gaussian kernel and is used to control the spatial influence range.

[0040] Sub-unit 1424 for constructing the correction factor matrix is ​​used to construct a diagonal matrix by using the node synthesis correction factors as diagonal elements, thus obtaining the correction factor matrix.

[0041] Secondly, this application also provides a method for intelligent monitoring and fault location of rural power distribution based on edge computing, including:

[0042] S1. Collect time-series data of each edge node of the power distribution network based on a unified protocol, and perform data cleaning, normalization and time alignment on the time-series data to obtain standard time-series data; wherein, the time-series data includes voltage, current, topology identifier and operating environment parameters;

[0043] S2. Based on standard time-series data, analyze the abnormal characteristics of voltage and current using a pre-set fault detection model to obtain fault characteristic data; wherein, the fault characteristic data is used to characterize the probability of each edge node failing.

[0044] S3. Based on the fault characteristic data and topology identifiers, calculate the probability distribution of the fault location based on the preset fault location model to obtain preliminary fault location data; wherein, the preliminary fault location data is used to characterize the distribution of possible fault locations;

[0045] S4. Based on the preliminary fault location data and operating environment parameters, correct the fault location results to obtain the fault location results; wherein, the fault location results are used to indicate the location area of ​​the fault.

[0046] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the system described in the first aspect.

[0047] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the system described in the first aspect.

[0048] The aforementioned edge computing-based intelligent monitoring and fault location system for rural power distribution networks utilizes an edge computing architecture to distribute data acquisition, feature analysis, preliminary location, and correction optimization to the network edge, enabling localized real-time processing of the monitoring and location process. This effectively overcomes the data transmission latency of traditional centralized processing, improving fault response speed. By integrating mathematical morphology, random forests, and topological fault propagation models, the system can accurately capture transient anomaly characteristics, reduce false alarms caused by load fluctuations, and achieve sensitive detection of early, minor faults. Further combining environmental risk and geospatial information for probabilistic correction enhances the adaptability of the location results to complex operating environments and improves location accuracy. The entire system adopts a lightweight model design, adapting to the limited computing resources of rural power grid edge nodes, achieving efficient, accurate, and adaptive intelligent fault monitoring and location. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a schematic diagram of the structure of a rural power grid distribution intelligent monitoring and fault location system based on edge computing in one embodiment;

[0051] Figure 2 This is a flowchart illustrating a method for intelligent monitoring and fault location of rural power distribution based on edge computing in one embodiment. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0053] refer to Figure 1 The document presents a schematic diagram of a smart monitoring and fault location system for rural power distribution based on edge computing, as provided in this application. The system includes:

[0054] The data acquisition and preprocessing module 11 is used to acquire time-series data from each edge node of the power distribution network based on a unified protocol, and to perform data cleaning, normalization and time-series alignment on the time-series data to obtain standard time-series data.

[0055] Optionally, the timing data includes voltage, current, topology identifier, and operating environment parameters.

[0056] Optionally, in the data acquisition stage, a data acquisition link is constructed based on a unified power system communication protocol. The communication protocol used has good interoperability and scalability, and is compatible with edge monitoring equipment produced by different manufacturers. Specifically, edge nodes include distribution transformer outgoing terminals, line branch boxes, user-side smart meters, and key line tower monitoring points, etc. Each edge node deploys a corresponding sensing and acquisition unit: voltage data is acquired through high-precision voltage sensors to ensure the capture of voltage surge details when a fault occurs; current data is acquired using Rogowski coil current sensors, which have the advantages of wide range and no insertion loss, and can adapt to the characteristics of large load fluctuations in rural power grids. The sampling frequency of voltage and current sensors is consistent; topology identification data is a unique code pre-assigned to each edge node and line segment, which contains information such as the line number to which the node belongs, voltage level, and geographical area, and is read in real time through the local storage unit of the edge node; operating environment parameters are acquired through integrated environmental sensors, covering parameters such as ambient temperature, relative humidity, rainfall, and wind speed, and the sampling frequency meets the time-series requirements for analyzing the impact of environmental factors on faults.

[0057] The data preprocessing stage sequentially performs data cleaning, normalization, and time-series alignment. For outliers caused by sensor faults, the 3σ criterion is used for identification and removal. This involves calculating the mean μ and standard deviation σ of each time-series data point, and classifying data points outside the range [μ-3σ, μ+3σ] as outliers, where μ is the mean and σ is the standard deviation. For missing values ​​generated during signal transmission, different supplementation strategies are employed based on the duration of the missing value. Shorter missing values ​​are supplemented using linear interpolation, while longer missing values ​​are marked as data failure segments and trigger the sensor self-test process. For high-frequency noise caused by electromagnetic interference, a wavelet threshold denoising algorithm is used. A suitable base wavelet is selected, and the corresponding decomposition level is set. The threshold is determined by calculating the noise variance, achieving noise filtering while preserving fault characteristic information.

[0058] The normalization process employs the min-max normalization algorithm, mapping time series data of different dimensions to the [0,1] interval. The specific calculation formula is as follows:

[0059]

[0060] in, The original data, This is the historical minimum value for this type of data. This represents the historical maximum value for this type of data. This is the result after normalization. and Data is obtained through statistics from the local historical database of edge nodes to ensure that the normalized data can be directly input into subsequent models. Time alignment is based on the local high-precision clock of the edge gateway, which is synchronized with GPS to ensure time accuracy. The time-series data of voltage, current, topology identifiers and environmental parameters are timestamped to align all data to the set time granularity, ensuring the consistency of different types of data in the time dimension.

[0061] The fault feature analysis module 12 is used to analyze the abnormal characteristics of voltage and current based on standard time series data and a preset fault detection model to obtain fault feature data.

[0062] Optionally, fault characteristic data is used to characterize the probability of each edge node failing.

[0063] Optionally, the preset fault detection model adopts a lightweight CNN-LSTM hybrid model, which takes into account both local feature extraction and temporal dependency modeling capabilities. After model quantization, the computational efficiency is significantly improved, which can meet the real-time inference requirements of edge nodes.

[0064] The model structure includes an input layer that receives normalized voltage and current time-series data; a convolutional layer employing multiple one-dimensional convolutional layers with set kernel sizes, using different output channel numbers and the ReLU activation function to extract local abnormal features in the voltage and current waveforms, such as voltage drops and current surges during fault occurrence; a pooling layer using max pooling, with appropriate kernel sizes and strides to reduce data dimensionality and computational complexity; an LSTM layer containing a set number of hidden units to capture the time-dependent relationships between voltage and current data and uncover parameter change trends before and after faults; and a fully connected layer containing a set number of neurons using the ReLU activation function. The output layer uses the Sigmoid activation function, outputting a single probability value between 0 and 1, which characterizes the probability of fault occurrence at the corresponding edge node.

[0065] The model training process requires constructing a training dataset based on historical fault data and normal operation data of the rural power grid. Historical fault data covers common fault types in rural power grids, including single-phase grounding faults, phase-to-phase short-circuit faults, and overload faults. Normal operation data is collected from the rural power grid during different seasons and load periods. To avoid sample imbalance, the SMOTE algorithm is used to oversample fault samples, ensuring a balanced ratio between fault and normal samples. During training, the Adam optimizer is used, with an initial learning rate and a learning rate decay strategy. The binary cross-entropy loss function is used. Training is stopped iteratively when the validation set loss shows no continuous decrease for multiple rounds, ensuring the model has high accuracy and low false alarm rate.

[0066] Specifically, firstly, standardized time-series data is divided into sliding windows according to a set time step. By setting an appropriate step size, continuous monitoring of the time-series data is achieved, avoiding missed fault detection. The voltage and current data of each sliding window are input into a trained lightweight CNN-LSTM model. The model outputs the fault probability value of each edge node through forward inference. Simultaneously, local feature vectors output by the model's convolutional layers are extracted and combined with statistical features of the voltage and current data, such as the voltage surge amplitude, current peak value, and voltage-current phase difference change at the time of fault occurrence, to construct fault feature data. When the fault probability value output by the model reaches a set threshold, the corresponding edge node is determined to have fault anomaly characteristics, and the timestamp of the anomaly, the edge node topology identifier, and the corresponding feature vector are recorded. If the fault probability value does not reach the set threshold, it is determined to be in normal operation, and no fault feature data is generated. In addition, to adapt to the interference caused by rural power grid load fluctuations and intermittent distributed power source access, the module has a built-in feature adaptive adjustment mechanism: it calculates the deviation between the current operating load and the historical average load in real time. When the deviation exceeds the set range, it automatically adjusts the sliding window size of the model input data and corrects the fault probability threshold, reducing the false alarm rate caused by load fluctuations and ensuring effective detection of early weak fault characteristics.

[0067] The preliminary fault location module 13 is used to calculate the probability distribution of the fault location based on the fault feature data and topology identifier, and obtain preliminary fault location data.

[0068] Optionally, preliminary fault location data is used to characterize the distribution of possible fault locations.

[0069] Optionally, based on the topology identifiers of each edge node, the rural power grid distribution line is abstracted as an undirected graph with edge nodes as vertices and line segments as edges. The adjacency matrix of the topology graph is generated through the line topology information stored locally by the edge nodes, and the elements of the adjacency matrix are... Represents vertices With vertex The connection relationships between them, where vertices Corresponding edge nodes ,vertex Corresponding edge nodes , This indicates that two nodes are directly connected, meaning there is a line segment. This indicates that there is no direct connection. Simultaneously, attribute parameters are assigned to each edge, i.e., line segment, including line length, conductor type, resistance, and reactance. These parameters are retrieved in real-time from the edge gateway's local database, providing line characteristic information for fault location.

[0070] The pre-defined fault location model employs a node classification model based on graph neural networks. This model can fully utilize the line topology information to achieve accurate fault location inference. The model structure includes: an input layer that receives fault feature data from each edge node and the adjacency matrix of the line topology graph. The fault feature data includes fault probability values, feature vectors, etc.; a graph convolutional layer using multiple GCN layers, by setting different output dimensions and combining the ReLU activation function, updates the node's feature representation by aggregating the feature information of each edge node's neighboring nodes, thus achieving the fusion of topology and fault features; a fully connected layer containing a set number of neurons, using the ReLU activation function; and an output layer using the Softmax activation function, outputting the fault probability value for each line segment, constituting the probability distribution of the fault location, i.e., the preliminary fault location data.

[0071] The model training process uses labeled historical fault data, with the labels being the line segment numbers where the faults actually occurred. The training uses the SGD optimizer with an appropriate learning rate and the cross-entropy loss function. The model is iteratively trained until its performance meets the set requirements, ensuring that the model has a high localization accuracy.

[0072] Specifically, firstly, the fault feature data output by the fault feature analysis module is associated with the corresponding edge node topology identifiers to determine the set of edge nodes with abnormal fault characteristics. Then, the feature data of this edge node set, the adjacency matrix of the line topology graph, and the line attribute parameters are input into the trained GNN fault localization model. The model aggregates neighbor node information through graph convolution operations, updates the feature representation of each edge node, and then calculates the fault probability of each line segment. Based on the fault probability values ​​of each line segment, preliminary fault localization data is formed, characterizing the distribution of line segments where faults may occur.

[0073] Furthermore, to enhance the model's adaptability to topology changes, the module incorporates a topology update monitoring mechanism: it monitors the topology identifier information uploaded by edge nodes in real time. When a change in topology identifier is detected, it automatically triggers the topology graph update process, regenerates the adjacency matrix, and loads the pre-trained topology adaptive model parameters, ensuring accurate calculation of the fault location probability distribution even in topology change scenarios. Simultaneously, considering the characteristics of rural power grids with numerous branches and complex structures, the model introduces a line length weighting factor to correct the fault probability of long line sections, improving the accuracy of fault location in long line sections.

[0074] The fault location correction module 14 is used to correct the fault location results based on the preliminary fault location data and operating environment parameters, and obtain the fault location results.

[0075] Optionally, the fault location results are used to indicate the location area of ​​the fault.

[0076] Optionally, the fault location correction module quantifies the impact of operating environment parameters on fault occurrence and propagation, establishes a correlation model between environmental parameters and fault location results, and improves location accuracy by correcting the probability distribution of preliminary fault location.

[0077] First, feature extraction is performed on the operating environment parameters, converting parameters such as ambient temperature, relative humidity, rainfall, and wind speed into quantitative feature indicators: temperature influence coefficient T, used to characterize the impact of ambient temperature on line insulation performance and fault probability; humidity influence coefficient H, used to characterize the impact of relative humidity on line insulation leakage current and fault probability; rainfall influence coefficient R, divided into different levels according to rainfall amount, with different levels corresponding to different influence coefficients, used to characterize the impact of rainfall on fault probability; wind speed influence coefficient W, divided into different levels according to wind speed amount, with different levels corresponding to different influence coefficients, used to characterize the impact of wind speed on line sway and phase-to-phase short circuit probability.

[0078] Based on the aforementioned environmental impact coefficients, a fault location correction model is constructed. The specific correction process is as follows: First, extract the line segments with the highest fault probability from the preliminary fault location data to form a candidate segment set; Second, obtain the operating environment parameters corresponding to each segment in the candidate segment set. If a segment does not have directly deployed sensors, the average environmental parameters of two adjacent edge nodes are used. The comprehensive environmental impact coefficient K for each candidate segment is calculated based on the obtained environmental parameters. K is the product of the temperature impact coefficient T, humidity impact coefficient H, rainfall impact coefficient R, and wind speed impact coefficient W; Third, the preliminary fault probability is corrected using the comprehensive environmental impact coefficient. The correction formula is:

[0079]

[0080] in, This represents the initial failure probability. This represents the comprehensive environmental impact coefficient for the current section. Let be the initial failure probability of the i-th candidate segment. For the first The comprehensive environmental impact coefficient of each candidate section, The corrected failure probability is calculated using this formula. The corrected failure probability is then normalized to ensure that the sum of the failure probabilities of all candidate segments is 1. The fourth step involves a secondary adjustment of the corrected failure probability using geospatial information: The geographical coordinates of candidate segments are obtained through GPS positioning information from edge nodes. It is then determined whether the segment is located in a low-lying area prone to flooding, a densely wooded area, or a wind-prone area. Different geographical areas correspond to different adjustment coefficients. The corrected failure probability is multiplied by the corresponding adjustment coefficient to achieve a deep fusion of geospatial and environmental characteristics.

[0081] Finally, the line segment with the highest fault probability after the second adjustment is selected as the fault location result. The specific location information of this segment is output, including its line number, the topology identifiers of the starting and ending edge nodes, the geographical coordinate range, and the confidence level of the fault occurrence, i.e., the highest fault probability after adjustment. Simultaneously, to ensure the reliability of the location results, the module has a built-in result verification mechanism: when the difference between the highest and second-highest fault probabilities reaches a set requirement, the location result is directly output; when the difference does not reach the set requirement, a supplementary monitoring process is triggered, controlling the edge nodes near the candidate segment to increase the sampling frequency, continuously collecting voltage and current data for a set duration, re-inputting it into the fault feature analysis module and the preliminary fault location module to obtain new preliminary fault location data, and performing correction calculations again until the difference between the two reaches the set requirement or the number of supplementary monitoring sessions reaches the set upper limit. If the difference requirement is still not met after the upper limit of supplementary monitoring sessions, the two segments with the highest fault probabilities are output as the joint location result. Through the above correction process, the interference of environmental and geographical factors on fault location can be effectively reduced, improving the accuracy of the final fault location result, meeting the needs of rapid emergency repair of rural power grid faults, and providing a clear basis for maintenance personnel to accurately carry out fault isolation and repair work.

[0082] The aforementioned edge computing-based intelligent monitoring and fault location system for rural power distribution networks utilizes an edge computing architecture to process data locally at the network edge, reducing communication bandwidth pressure and decision latency, thus enabling rapid fault response. The system effectively captures transient voltage and current fluctuations and early weak features through advanced feature extraction and lightweight machine learning model fusion analysis, improving the sensitivity and accuracy of fault detection and reducing false alarms. It dynamically constructs a network model using topology identifiers, enabling fault location to adapt to changes in network structure. Furthermore, it incorporates environmental temperature, humidity, and geospatial information for probabilistic correction, overcoming the limitations of traditional methods that ignore on-site factors and improving location accuracy. The entire process employs optimized algorithms and lightweight design, fully accommodating the limited resources of rural power grid edge nodes, achieving efficient, accurate, and adaptive intelligent monitoring and fault location.

[0083] In one embodiment, the fault feature analysis module 12 includes:

[0084] The feature extraction subunit 121 is used to extract features from voltage and current sequences in standard time series data based on mathematical morphology gradient operators, capture transient change information of waveforms, and obtain gradient feature vectors.

[0085] Optionally, the mathematical morphological gradient operator can effectively enhance the abrupt changes in a signal and suppress noise interference in stationary segments by performing morphological operations on the signal. Specifically, the standardized voltage and current time-series data are first preprocessed to ensure the continuity and integrity of the data sequence. Then, appropriate structuring elements are selected to perform morphological gradient operations on the voltage and current sequences respectively. The core operation logic of the mathematical morphological gradient is the difference between dilation and erosion operations. That is, dilation is performed on the signal using structuring elements to obtain an dilated signal, and erosion is performed on the same signal to obtain an eroded signal. The difference between the dilated signal and the eroded signal yields the morphological gradient signal. The peak value of this gradient signal corresponds to the transient change position of the voltage or current waveform, and the peak amplitude reflects the intensity of the change. Arranging the gradient signal values ​​at different times in chronological order constitutes a gradient feature vector. This vector completely preserves the transient change information of the voltage and current waveforms in the time dimension, providing basic feature support for subsequent fault diagnosis.

[0086] The energy entropy calculation subunit 122 is used to calculate the energy entropy of the gradient feature vector within a preset time window to obtain the entropy feature value.

[0087] Optionally, energy entropy is used to quantify the complexity and abrupt change intensity of the signal.

[0088] Optionally, since the transient changes in voltage and current waveforms during rural power grid faults significantly increase signal complexity, energy entropy, as a quantitative indicator based on information theory, can effectively characterize the disorder of signal energy distribution and thus reflect the waveform mutation characteristics caused by the fault. Specifically, a preset time window is first set, the length of which is determined according to the duration of the transient process of the rural power grid fault, ensuring complete coverage of the gradient feature information of the transient fault stage. Then, the gradient feature vector is slidably divided according to the preset time window to obtain multiple gradient sub-vectors within the window. Energy calculation is performed on the gradient sub-vector within each window. First, the square value of each element in the gradient sub-vector (i.e., the energy value at a single moment) is calculated, and then the energy values ​​at all moments within the window are summed to obtain the total energy of the window. Then, the energy values ​​at each moment within the window are normalized to obtain the energy distribution probability, i.e., the ratio of the energy value at a single moment to the total energy of the window. Finally, the energy entropy is calculated based on the energy distribution probability. The formula for calculating energy entropy is:

[0089]

[0090] in, For energy entropy, The number of sampling points within a preset time window. For the first The energy distribution probability at each sampling moment. The larger the entropy characteristic value calculated by this formula, the more disordered the energy distribution of the gradient characteristic vector within the window, and the greater the abrupt change intensity of the corresponding voltage or current waveform and the higher the signal complexity. Conversely, it indicates that the waveform is relatively stable and there are no obvious fault transient characteristics.

[0091] The initial fault score subunit 123 is used to input the gradient feature vector and entropy feature value into the preset lightweight random forest model for classification reasoning and output the initial fault score of each node.

[0092] Optionally, a lightweight random forest model is chosen as the classification inference model because it possesses strong feature fusion and anti-interference capabilities. Furthermore, through lightweight optimizations such as pruning and reducing the number of decision trees, the computational load is significantly reduced, making it suitable for the limited computing resources of rural power grid edge nodes. Model construction process: First, a training dataset is constructed. The samples in the training dataset consist of gradient feature vectors, entropy feature values, and corresponding fault labels corresponding to historical operating data. The fault labels are labeled as faulty or normal states based on the actual operating status. Then, the lightweight random forest model is trained. By setting hyperparameters such as the number of decision trees, decision tree depth, and feature sampling ratio, the model learns the mapping relationship between gradient feature vectors, entropy feature values, and fault states. Cross-validation is used to optimize hyperparameters during training to ensure stable classification performance under different operating scenarios in the rural power grid. During classification inference, the gradient feature vectors and entropy feature values ​​are concatenated to form a fused feature vector, which is then input into the trained lightweight random forest model. The model performs parallel inference on the fused feature vector through each decision tree to obtain the classification result of each decision tree. Then, it integrates the classification results of all decision trees through a voting mechanism and outputs the initial fault score of each edge node. This score is used to initially quantify the probability of edge node failure.

[0093] The fault feature subunit 124 is used to normalize the initial fault score using the Sigmoid activation function to obtain fault feature data.

[0094] Optionally, since the initial fault score has a variable numerical range and cannot be directly used as input for subsequent fault localization models, the Sigmoid activation function can be used to map the initial fault score to the [0,1] interval, giving the output fault feature data a unified quantification standard, which facilitates feature fusion and probability calculation in subsequent modules. The formula for calculating the Sigmoid activation function is:

[0095]

[0096] in, The fault characteristic data is normalized. An initial fault score is assigned. After calculation using this formula, the closer the fault characteristic data value is to 1, the higher the probability of a fault occurring at the corresponding edge node; the closer the value is to 0, the higher the probability that the edge node is operating normally. Furthermore, to adapt to interference caused by rural power grid load fluctuations and intermittent distributed power source access, the fault characteristic subunit incorporates a scoring correction mechanism: it acquires the current operating load data of the edge node in real time, calculates the deviation between the current load and the historical average load, and dynamically corrects the fault characteristic data output by the Sigmoid function when the deviation exceeds a set range. The correction logic involves adjusting the correction coefficient based on the load deviation magnitude, fine-tuning the fault characteristic data through the correction coefficient, reducing the risk of misjudgment of faults caused by load fluctuations, and ensuring that the fault characteristic data accurately reflects the actual operating status of the edge node.

[0097] In one embodiment, the preliminary fault location module 13 includes:

[0098] The adjacency matrix construction sub-unit 131 is used to construct the adjacency matrix of the power distribution network based on the topology identifier.

[0099] Optionally, the topology identifier, as a unique code pre-assigned to each edge node and line segment, contains key information such as the line number to which the node belongs, voltage level, geographical region, and connection relationships between nodes. The adjacency matrix construction sub-unit determines the connection status of each edge node by parsing this topology identifier information. Specifically, firstly, all edge nodes are numbered and sorted. The dimension of the adjacency matrix is ​​determined based on the number of edge nodes. If there are N edge nodes in the distribution network, an N×N dimensional adjacency matrix A is constructed, where the rows and columns of the matrix correspond to the sorted edge nodes. Then, the topology identifier of each edge node is traversed to extract the connection relationship information between nodes contained therein. If the edge node... With edge nodes If there is a direct connection, i.e., a line segment exists, then the elements at the corresponding positions in the adjacency matrix will be... Set to 1; if edge node With edge nodes If there is no direct connection, then the element will be... Set to 0. Since rural power grid distribution lines are undirected connections, the adjacency matrix satisfies symmetry, i.e. The completed adjacency matrix is ​​stored locally through the edge gateway and can be dynamically adjusted according to updates to the topology identifiers to ensure that the matrix can reflect the actual topology of the power distribution network in real time.

[0100] The fault state vector construction sub-unit 132 is used to construct the fault state vector of each node based on the fault characteristic data.

[0101] Optionally, the fault feature data is normalized data output by the fault feature analysis module, with a numerical range of [0,1], directly reflecting the probability of fault occurrence at the corresponding edge node. Specifically, the dimension of the fault state vector is first determined, and its dimension is consistent with the number of edge nodes in the distribution network, i.e., the fault state vector... This is an N×1 dimensional column vector (N being the number of edge nodes). Then, the edge nodes in the sub-unit are sorted according to their numbers based on the adjacency matrix, and the fault feature data corresponding to each edge node is sequentially used as the fault state vector. The elements, namely The corresponding number is The fault feature data of the edge nodes. If an edge node has no valid fault feature data (e.g., no fault anomaly features were detected), then the corresponding vector elements will be... Setting it to 0 indicates that the node currently shows no signs of failure. The constructed fault state vector strictly corresponds to the node number in the adjacency matrix, ensuring the validity and accuracy of subsequent matrix operations. At the same time, the vector data is updated in real time as the fault feature data is updated, ensuring the real-time representation of the fault state.

[0102] The fault propagation matrix calculation subunit 133 is used to calculate the fault propagation matrix based on the adjacency matrix and the fault state vector; wherein, the fault propagation matrix is ​​used to characterize the propagation intensity of fault features in the topological network; the expression of the fault propagation matrix is:

[0103]

[0104] in, This is the fault propagation matrix. It is an adjacency matrix. The Hadamard product represents the matrix multiplication, used for element-wise multiplication. For the fault state vector, It is a vector of all 1s. express transpose, This represents the absolute value operation. To prevent extremely small positive numbers with a denominator of zero.

[0105] In the formula, Given an N×N dimensional fault propagation matrix, its elements are... Characterizing faults from edge nodes Propagate to edge nodes The strength; It is an N×N dimensional adjacency matrix used to limit the scope of fault propagation (propagation is possible only between directly connected nodes). This represents the Hadamard product operation of matrices. The operation rule is to multiply corresponding elements of two matrices of the same dimension. The Hadamard product allows the fault propagation matrix to retain only the fault propagation intensity between directly connected nodes, while the propagation intensity of non-directly connected nodes is set to 0. It is an N×1 dimensional fault state vector, representing the probability of each node's own failure. It is an N×1 dimensional column vector of all 1s; Represents the fault state vector The transpose of is a 1×N dimensional row vector; Indicates all Column vectors and Matrix multiplication of transposed row vectors results in an N×N dimensional matrix; This indicates that the absolute value operation is performed on each element in the matrix to ensure that the fault propagation strength is non-negative; To prevent extremely small positive numbers with a denominator of zero from being used in subsequent calculations, their values ​​are much smaller than 1, so as not to affect the quantification results of fault propagation strength. If the node With nodes No direct connection ( ),but If node With nodes Direct connection ( ),but This accurately reflects the intensity of fault propagation between directly connected nodes.

[0106] The preliminary fault location subunit 134 is used to perform weighted summation and normalization on the row vectors of the fault propagation matrix to generate the fault probability distribution of each line section and obtain preliminary fault location data.

[0107] Optionally, each line segment in the distribution network corresponds to two directly connected edge nodes (i.e., non-zero elements in the adjacency matrix). By weighted summing of the row vectors of the fault propagation matrix, the fault propagation intensity information of each node can be aggregated, thereby quantifying the fault probability of the corresponding line segment. In implementation, the weighting factors for the weighted summation are first determined. These weighting factors are based on the attribute parameters of the line segment. Parameters such as line resistance, reactance, and length are selected to construct a weight calculation model. The larger the line resistance and reactance, and the longer the length, the larger the corresponding weighting factor (the probability of a fault occurring in such line segments is relatively higher). After normalizing the line attribute parameters through the weight calculation model, the weight value of each line segment is obtained. (corresponding node) and (the line section between).

[0108] Subsequently, the fault propagation matrix was analyzed. We perform a weighted summation of the row vectors of the fault propagation matrix, and then... Line, its elements Corresponding node To node The fault propagation strength is used as the weighting factor for all elements in the row and their corresponding line segments. Multiply and then sum to get the nodes. Overall fault propagation intensity of all associated line segments Since each line segment corresponds to two nodes. Its overall fault propagation strength will be affected by the nodes. and nodes The calculation is performed separately, so the average of the two calculation results for the same line segment is needed to obtain the final comprehensive fault propagation intensity for that line segment.

[0109] Finally, the final comprehensive fault propagation intensity of all line sections is normalized. The min-max normalization algorithm is used to map it to the [0,1] interval to obtain the fault probability of each line section. The fault probabilities of all line sections constitute the probability distribution of the fault location, which is the preliminary fault location data. This data clearly represents the line sections where faults may occur and the corresponding probability.

[0110] In one embodiment, the fault location and correction module 14 includes:

[0111] The environmental risk coefficient subunit 141 is used to extract the ambient temperature and ambient humidity from the operating environment parameters, and query the environmental risk coefficient corresponding to the current environmental conditions based on the preset historical fault statistics table.

[0112] Optionally, the occurrence of rural power grid faults is significantly correlated with ambient temperature and humidity. High temperature and high humidity environments can easily lead to a decrease in the insulation performance of lines, increasing the probability of fault occurrence. Therefore, these two core environmental parameters are selected as the basis for risk quantification. The preset historical fault statistics table is constructed based on historical fault data and corresponding environmental data from long-term operation of the rural power grid. In the construction process, the historical fault data is first classified and organized, and key information such as ambient temperature, ambient humidity, fault location, and fault type at the time of each fault is extracted. Then, the ambient temperature and ambient humidity are divided into several continuous intervals. Using the temperature interval and humidity interval as two-dimensional indexes, the frequency of fault occurrence, fault rate, and fault severity in each two-dimensional interval are statistically analyzed. Finally, the environmental risk coefficient corresponding to each two-dimensional environmental interval is quantified based on the statistical results. The risk coefficient ranges from [1, K_max], where a larger value indicates a higher risk of fault occurrence under the environmental conditions, and a value of 1 indicates that the environmental conditions have no significant impact on the occurrence of faults.

[0113] Specifically, firstly, the current ambient temperature and humidity data are extracted in real time from the operating environment parameters. The extracted data is then validated, and abnormal data (such as data exceeding the sensor's measurement range) is removed. If the data is invalid, the average of the environmental parameters from the most recent three sampling periods is used as a replacement. Next, based on preset temperature and humidity ranges, a two-dimensional interval is determined to which the current ambient temperature and humidity belong. This two-dimensional interval is used as an index to query the historical fault statistics table and obtain the corresponding environmental risk coefficient. If the current environmental parameters happen to fall within the critical values ​​of the two intervals, the corresponding environmental risk coefficient is calculated using linear interpolation. The calculation formula is as follows:

[0114]

[0115] in, The environmental risk coefficient under current environmental conditions. The risk coefficient of the adjacent left interval. The risk coefficient of the adjacent right interval. This is the difference between the current environmental parameters and the critical value of the left interval. This represents the difference between the current environmental parameters and the critical value of the interval on the right. This method ensures that the environmental risk coefficient accurately matches the current environmental conditions, providing a reliable basis for risk quantification for subsequent corrections.

[0116] Correction factor construction subunit 142 is used to construct a correction factor matrix based on the environmental risk coefficient.

[0117] Optionally, the dimension of the correction factor matrix is ​​consistent with the number of line segments in the preliminary fault location data. If the preliminary fault location data contains M line segments, then an M×M dimensional correction factor matrix F is constructed. Since the operating environment of each line segment is different (e.g., the temperature and humidity of different segments may vary due to different geographical locations), a corresponding environmental risk coefficient needs to be matched for each line segment. Therefore, the correction factor matrix adopts a diagonal matrix form, that is, only the diagonal elements are non-zero, and the off-diagonal elements are all 0, ensuring that the fault probability of each line segment is corrected only by its own corresponding environmental risk coefficient, avoiding mutual interference between environmental risks of different segments.

[0118] Specifically, firstly, the identification information of all line segments in the preliminary fault location data is obtained, and a one-to-one correspondence is established between the line segments and the diagonal elements of the correction factor matrix, that is, the matrix... Line number Column elements Corresponding to the The correction factor for each line segment. Then, the environmental risk coefficients corresponding to each line segment obtained from the environmental risk coefficient sub-unit are used as the values ​​of the diagonal elements, i.e. ,in For the first The environmental risk coefficient for each line segment. If a line segment has no directly corresponding environmental parameters (e.g., no environmental sensors are deployed), the average environmental risk coefficient of the edge nodes at both ends of that line segment is used as its correction factor, ensuring that each line segment has a corresponding correction factor. The constructed correction factor matrix must strictly correspond to the order of the line segments in the preliminary fault location data to avoid correction failure due to disordered order. Simultaneously, the matrix is ​​updated in real time as the environmental risk coefficient is updated to ensure that the correction factors reflect the risk characteristics under the current environmental conditions.

[0119] The fault probability correction subunit 143 is used to perform a dot product operation on the probability distribution and correction factor matrix in the preliminary fault location data to obtain the weighted corrected fault probability.

[0120] Specifically, the fault probability distribution in the preliminary fault location data is first converted into an M×1 dimensional column vector P, where the vector elements... Corresponding to the The initial fault probability of each line segment is calculated, where M is the number of line segments. Then, this fault probability column vector is multiplied by the M×M dimensional correction factor matrix F constructed by the correction factor construction subunit. The specific logic of the dot product operation is the multiplication of matrix F and vector P, resulting in an M×1 dimensional column vector. ,Right now:

[0121]

[0122] in, The elements are the weighted and corrected fault probability vector. Corresponding to the The corrected fault probability of each line segment This calculation process, through the diagonal characteristics of the correction factor matrix, achieves precise weighting of the initial failure probability and its own environmental risk coefficient for each line segment. The higher the environmental risk coefficient of the line segment, the more significantly its failure probability is amplified, thereby highlighting the failure possibility of high environmental risk segments and reducing the impact of high initial failure probability caused by accidental factors in low environmental risk segments.

[0123] To ensure the reasonableness of the corrected fault probabilities, the results must be validated after the dot product operation. If the corrected fault probability of a certain line segment is greater than 1, it is corrected to 1 (the maximum fault probability is 1); if the corrected fault probability is less than 0, it is corrected to 0 (the minimum fault probability is 0). Furthermore, if multiple line segments have the same corrected fault probability, all of which are the maximum value, the identification information of these segments must be recorded to provide a basis for subsequent fault location determination.

[0124] The fault location determination subunit 144 is used to select the area with the largest value in the weighted and corrected fault probability as the final fault point and output the fault location result.

[0125] Specifically, firstly, the weighted and corrected fault probability vector... The data is sorted, and the line segment corresponding to the element with the largest value is extracted and identified as the final fault point. If multiple elements with the same maximum fault probability exist after sorting (i.e., multiple line segments have the same and maximum corrected fault probability), a second screening is performed based on the topological relationships and historical fault frequencies of these line segments: line segments directly connected to the edge nodes corresponding to the fault feature data are selected first; if they still cannot be distinguished, line segments with higher historical fault frequencies are selected as the final fault point; if the historical fault frequencies are the same, all these line segments are considered as joint fault points.

[0126] The final fault location result includes detailed location information, specifically: the line segment identifier corresponding to the final fault point, the line number, the starting edge node topology identifier, the ending edge node topology identifier, the geographical coordinate range of the line segment (calculated by interpolation based on the GPS positioning information of the edge nodes), and the corresponding weighted corrected fault probability (i.e., the confidence level of the fault occurrence). Simultaneously, to ensure the reliability of the location results, the sub-unit has a built-in result feedback verification mechanism: the output fault location result is synchronized to the local storage unit of the edge gateway and compared with the fault characteristics corresponding to subsequently collected operational data. If subsequent data does not detect the persistent fault characteristics of the fault point, a recalibration process is triggered, re-executing the environmental risk coefficient extraction, correction factor matrix construction, and fault probability correction processes to ensure the accuracy and timeliness of the location results, providing reliable technical support for the rapid isolation and repair of rural power grid faults.

[0127] In one embodiment, the fault preliminary location subunit 134 includes:

[0128] The normalization subunit 1341 is used to normalize each row vector of the fault propagation matrix using the L1 norm to obtain the node fault probability value of each node.

[0129] Optionally, the row vectors of the fault propagation matrix represent the fault propagation intensity from the corresponding node to other nodes. Since their numerical range is not uniform, they cannot be directly used as fault probabilities. L1 norm normalization can map the row vectors to the [0,1] interval, thus converting fault propagation intensity into probability values. The core logic of L1 norm normalization is to calculate the sum of the absolute values ​​of all elements in the row vector, and then compare each element with this sum to ensure that the sum of all elements in the row vector is 1 after normalization, conforming to the basic characteristics of a probability distribution. The specific calculation formula is as follows:

[0130]

[0131] in, The normalized fault propagation matrix is ​​the first... Line number The elements of a column represent nodes. To the node The probability component of propagation failure; For the fault propagation matrix, the first... Line number The original elements of a column, i.e., nodes To the node The intensity of fault propagation; The number of edge nodes. For the fault propagation matrix, the first... The L1 norm of a row vector is the sum of the absolute values ​​of all elements in that row. Specifically, the L1 norm normalization operation is performed sequentially on each row vector of the fault propagation matrix to obtain a normalized inter-node fault propagation probability matrix. Then, the elements of each row vector in this matrix are summed (theoretically, the summation result should be 1) to verify the effectiveness of the normalization. If the error of the summation result of a row deviating from 1 exceeds a set threshold, the normalization operation is performed again. Finally, the diagonal elements corresponding to each node in the normalized matrix are extracted (i.e.,... This value serves as the node failure probability value for each node, representing the probability of the corresponding node itself failing.

[0132] The line segment fault probability subunit 1342 is used to calculate the fault probability of each line segment based on the topological connection relationship of the power distribution network, and obtain the initial probability of the line segment.

[0133] Optionally, each line segment in the distribution network corresponds to two directly connected edge nodes (i.e., in the adjacency matrix). The failure probability of a line segment is directly related to the failure probabilities of the nodes at both ends. Therefore, the node failure probabilities need to be mapped to the line segment through topological connections. Specifically, first, the adjacency matrix constructed by the adjacency matrix construction sub-unit is retrieved, and the node pairs corresponding to all non-zero elements are extracted. Each node pair uniquely corresponds to a line segment, establishing a one-to-one mapping relationship between node pairs and line segments. Subsequently, based on the node failure probability values ​​at both ends of the node pair, the failure probability of the corresponding line segment is calculated. A weighted summation algorithm is used to convert the node probability to the segment probability. The specific calculation formula is as follows:

[0134]

[0135] in, For nodes and The initial probability of the line segment between the line segments; For nodes The node failure probability value; For nodes The node failure probability value; This is a weighting coefficient, and its value is determined based on the topological attributes of the line segment. If the line segment is close to the power source, then... The value is too large, and it is too close to the load end. If the initial probability value is too small, the weighting coefficient can be adjusted to adapt to the fault propagation characteristics of the radial topology of the rural power grid. If a line segment has branch nodes, the fault probability value of the branch nodes needs to be added in addition. The initial probability of the line segment is recalculated using a weighted average algorithm to ensure that the probability calculation of the branch line segment fully considers the fault states of all related nodes. After the calculation is completed, the initial probabilities of all line segments are sorted by line number to form a set of initial probabilities of line segments.

[0136] The candidate fault screening subunit 1343 is used to filter the initial probability of the line segment by a preset dynamic truncation threshold, remove probability segments below the preset threshold, and obtain a set of candidate faults.

[0137] Optionally, a dynamic truncation threshold can be used instead of a fixed threshold to adapt to the different fault probability distributions under different operating scenarios of rural power grids, avoiding missed or incorrect screenings caused by a fixed threshold. The dynamic truncation threshold is set based on the statistical characteristics of the initial probability set of the line segment. The specific setting process is as follows: first, calculate the mean and standard deviation of the initial probability set of the line segment, and then determine the dynamic truncation threshold based on the mean and standard deviation. The calculation formula is:

[0138]

[0139] in, The mean, Standard deviation, To adjust the coefficient, its value is dynamically adjusted according to the frequency of rural power grid failures, with adjustments made during periods of high failure incidence. The value is too small, indicating a low incidence of failure. The value is set relatively large to ensure that the threshold accurately matches the current fault occurrence situation. During implementation, each element in the initial probability set of a line segment is compared with the dynamic truncation threshold T. If the initial probability of a certain line segment is too large... If so, the line section is retained and included in the candidate fault set; if , it is determined that the possibility of a fault occurring in this line section is extremely low, and it is excluded. At the same time, to avoid a too high dynamic truncation threshold caused by extreme outliers, it is necessary to perform outlier detection on the initial probability set of the line section, use the 3σ criterion to exclude extreme large values in the probability set, and then recalculate the mean and standard deviation to determine the dynamic truncation threshold. After the screening is completed, record the identification information, initial probability, and corresponding geographical coordinates of each line section in the candidate fault set to provide data support for subsequent clustering processing.

[0140] The preliminary fault location subunit 1344 is used to perform aggregation processing on the sections in the candidate fault set through a density-based spatial clustering algorithm, and output the cluster with the highest probability density as the preliminary fault location data.

[0141] Optionally, the density-based spatial clustering algorithm (DBSCAN) has the advantages of not requiring pre-specification of the number of clusters and being able to identify clusters of any shape. It is suitable for scenarios where the rural power grid lines are scattered and the shape of the fault area is not fixed, and can effectively aggregate adjacent high-probability fault sections and accurately lock the core fault area. Specifically, first extract the spatial features of each line section in the candidate fault set, and use the midpoint geographical coordinates of the line section (interpolated and calculated based on the GPS coordinates of the starting and ending edge nodes) as the spatial feature vector for clustering. Subsequently, set the core parameters of the DBSCAN algorithm: (neighborhood radius) and MinPts (the minimum number of neighborhood points of a core point), where is determined according to the average spacing of the rural power grid lines to ensure that adjacent fault line sections can be划入 the same neighborhood; MinPts is dynamically adjusted according to the scale of the candidate fault set. The larger the set scale, the larger the value of MinPts, to avoid generating too many small clusters.

[0142] The clustering process is specifically as follows: traverse each line section in the candidate fault set, use the spatial feature vector of this section as the center, and find the number of other line sections in the neighborhood. If the number ≥ MinPts, mark this section as a core point and include all sections in its neighborhood in the same cluster; if the number < MinPts but within the neighborhood of a core point, mark it as a boundary point and归入 the cluster corresponding to the core point; if it is neither a core point nor a boundary point, mark it as a noise point and exclude it. After traversing all sections, calculate the probability density of each cluster. The calculation method of the probability density is the ratio of the sum of the initial probabilities of all line sections in the cluster to the number of sections in the cluster, and this value represents the comprehensive possibility of a fault occurring in the cluster area. Finally, select the cluster with the highest probability density as the core fault area, and output all the line section information in this cluster as the preliminary fault location data. The output content includes the identification, initial probability, geographical coordinate range of the line sections in the cluster, and the probability density of the cluster.

[0143] In one embodiment, the correction factor construction subunit 142 includes:

[0144] Geographic coordinate matrix construction sub-unit 1421 is used to obtain the geographical coordinates of each edge node in the power distribution network and construct the geographic coordinate matrix.

[0145] Optionally, high-precision geographic coordinates (latitude and longitude coordinates) are obtained using the GPS positioning module built into the edge nodes, with the positioning error controlled within a set range. Specifically, firstly, geographic coordinate information is extracted from the local storage unit or real-time uploaded data of each edge node. The extracted coordinate data is validated, and data with positioning failures or errors exceeding the standard are discarded. If the coordinate data of an edge node is invalid, approximate coordinates are calculated by interpolating the coordinates of its adjacent edge nodes. Subsequently, a geographic coordinate matrix G is constructed according to the edge node numbers in the adjacency matrix construction sub-unit. The matrix dimension is N×2 (N is the number of edge nodes), where the matrix number is... The element in the first column of the row is a node. The longitude coordinates, the first The element in the second column of the row is a node. The latitude coordinates, i.e. , and They are numbered as follows The longitude and latitude of the edge nodes are determined. The constructed geographic coordinate matrix strictly corresponds to the node numbers in the adjacency matrix and fault state vector, ensuring consistency between subsequent spatial distance calculations and the topology.

[0146] The spatial distance matrix calculation subunit 1422 is used to calculate the spatial distance between nodes using the Euclidean distance formula on the geographic coordinate matrix, thus obtaining the spatial distance matrix.

[0147] Optionally, Euclidean distance can accurately quantify the straight-line distance between two nodes in geographic space, providing basic data for subsequent spatial weight calculations and is suitable for distance characterization of the spatial distribution of rural power grid nodes. During implementation, based on the geographic coordinate matrix G, for any two edge nodes... and ( To calculate spatial distance, the formula for Euclidean distance is:

[0148]

[0149] in, For nodes With nodes The Euclidean distance between them and They are nodes and nodes geographic coordinate vector ( ), , They are nodes Longitude and latitude , They are nodes The longitude and latitude of the nodes are used. To eliminate the distance quantization bias caused by latitude and longitude units, the latitude and longitude coordinates need to be converted to Cartesian coordinates before calculation (e.g., using Gauss-Kruger projection). After performing Euclidean distance calculations on all node pairs sequentially, an N×N spatial distance matrix D is constructed, with matrix elements... , characterizing nodes With nodes The spatial straight-line distance between them. Since spatial distance is symmetric, the spatial distance matrix satisfies... And diagonal elements (The distance between a node and itself is 0).

[0150] The comprehensive correction factor calculation subunit 1423 is used to calculate the Gaussian radial basis function value as spatial weight based on the spatial distance matrix, and combine it with the environmental risk coefficient to generate the node comprehensive correction factor;

[0151] The expression for the comprehensive correction factor is as follows:

[0152]

[0153] In the formula, Represents a node The comprehensive correction factor, Represents a node Environmental risk coefficient and Representing nodes respectively and nodes The geographic coordinate vector, Represents a node The set of neighboring nodes, This represents the bandwidth parameter of the Gaussian kernel, used to control the spatial influence range.

[0154] Optionally, spatial weights are used to quantify the spatial influence of neighboring nodes on the failure probability of the current node; the closer the neighboring nodes are, the greater their influence. Combined with the environmental risk coefficient, environmental and spatial factors can be integrated, allowing the correction factor to more comprehensively reflect the influencing conditions of failure occurrence. During implementation, the set of neighboring nodes N(i) for each node is first determined based on the adjacency matrix. The set of neighboring nodes N(i) is defined as the set of nodes with respect to each other. The set of directly connected edge nodes (i.e., in the adjacency matrix) All nodes This ensures that spatial weights only consider nodes that are directly related in the topology, avoiding interference from irrelevant nodes.

[0155] Then, the spatial weight summation of each node is calculated. The spatial weights are calculated using the Gaussian radial basis function, which has the characteristics of good smoothness and controllable local influence range, and can accurately characterize the attenuation characteristics of spatial distance on weights. The specific calculation logic is as follows: for each node... Each neighbor node Calculate the Gaussian radial basis function values, and then sum the function values ​​over all neighboring nodes. Finally, combine this with the environmental risk coefficient to generate a node comprehensive correction factor, as shown in the above expression: Represents a node The comprehensive correction factor is used to quantize nodes. Fault probability correction strength for associated line sections; Represents a node The environmental risk coefficient is calculated from the environmental risk coefficient sub-unit and characterizes the node. The impact of environmental conditions on the risk of failure; and Representing nodes respectively and nodes The geographic coordinate vector, i.e. ; Represents a node With nodes The Euclidean distance between them is obtained by calculating the sub-units from the spatial distance matrix; Represents a node The set of neighboring nodes is determined based on the adjacency matrix; The bandwidth parameter of the Gaussian kernel is used to control the spatial influence range. Its value is determined based on the average spacing of the rural power grid lines; the larger the spacing, the better. A larger value ensures that the influence range of the spatial weight matches the actual line distribution; The values ​​of the Gaussian radial basis function, i.e., the nodes. For nodes The spatial weights are such that the greater the spatial distance, the smaller the weight value, which conforms to the law that the impact of a fault decreases with distance.

[0156] Specifically, for each node First, extract the set of its neighbor nodes from the adjacency matrix. Then obtain the nodes from the spatial distance matrix. With each neighbor node Euclidean distance Substitute the values ​​into the Gaussian radial basis function formula to calculate each neighbor node. The corresponding spatial weights; sum the spatial weights of all neighboring nodes to obtain the sum of spatial weights; then compare this sum with the node's spatial weights. Environmental risk coefficient Multiply to get the node Comprehensive correction factor If node If there are no neighboring nodes (such as an isolated monitoring node), the sum of spatial weights is set to 1. This ensures the effectiveness of the comprehensive correction factor.

[0157] Sub-unit 1424 for constructing the correction factor matrix is ​​used to construct a diagonal matrix by using the node synthesis correction factors as diagonal elements, thus obtaining the correction factor matrix.

[0158] Optionally, the correction factor matrix is ​​in diagonal form, which ensures that the fault probability of each line segment is corrected only by the comprehensive correction factor of its associated nodes, avoiding cross-interference between different line segments, and simplifying the matrix operation process to meet the lightweight computing requirements of edge nodes. Specifically, the dimension of the correction factor matrix is ​​first determined, and its dimension is consistent with the number of line segments in the preliminary fault location data. If the preliminary fault location data contains M line segments, then an M×M dimensional correction factor matrix F is constructed.

[0159] Subsequently, the correlation between line segments and node comprehensive correction factors was established, with each line segment corresponding to two directly connected edge nodes. Take the combined correction factor of these two nodes. and The mean value is used as the comprehensive correction factor for the corresponding line segment, i.e., the line segment Corresponding comprehensive correction factor , where nodes and For line section The two edge nodes are considered. If a line segment is associated with only one edge node (such as a terminal line segment), the comprehensive correction factor of that node is directly taken as the comprehensive correction factor of the line segment. If a line segment does not have a directly corresponding node comprehensive correction factor (such as invalid coordinates of the nodes at both ends of the line segment and no adjacent nodes), the average environmental risk coefficient of the area to which the line belongs is used as the base value, and an alternative comprehensive correction factor is calculated by combining the average spatial weight of the area, ensuring that each line segment has a corresponding correction basis.

[0160] Finally, a diagonal matrix is ​​constructed, and the comprehensive correction factor corresponding to each line segment is sequentially used as the diagonal element of the correction factor matrix F, i.e., the matrix... Line number Column elements All off-diagonal elements are set to 0. The completed correction factor matrix must strictly correspond to the line segment order in the preliminary fault location data to avoid correction failure due to disordered order. Simultaneously, the matrix is ​​updated in real time as the node comprehensive correction factor is updated. Specifically, when changes in environmental parameters lead to updates in the environmental risk coefficient, or when changes in the topology cause changes in the neighbor node set or spatial distance, the node comprehensive correction factor and correction factor matrix are automatically recalculated. This ensures that the correction factors can reflect the current environmental conditions and topology in real time, providing dynamic support for accurate fault probability correction.

[0161] The aforementioned edge computing-based intelligent monitoring and fault location system for rural power distribution integrates data acquisition, feature analysis, preliminary location, and intelligent correction into a hierarchical and collaborative edge computing architecture, achieving localized and intelligent fault handling throughout the entire process. The system utilizes mathematical morphological gradients and energy entropy to accurately capture early, subtle fault characteristics, combined with a lightweight machine learning model for real-time analysis, effectively overcoming the high false alarm rate problem of traditional threshold methods. By constructing a fault propagation matrix model, it intelligently propagates and aggregates node anomaly scores within the topology network, achieving highly robust preliminary probabilistic location. Furthermore, it introduces environmental risk coefficients and geospatial weights for correction, improving the accuracy of location results under complex operating environments. By reducing data transmission bandwidth requirements and central processing latency, the system can adapt to the resource-constrained edge node environment of rural power grids, enabling rapid and accurate fault perception and location, and improving the reliability of rural power supply.

[0162] Based on the same inventive concept, this application also provides a method for implementing the aforementioned edge computing-based intelligent monitoring and fault location system for rural power distribution. The solution provided by this method is similar to the implementation scheme described in the above system. Therefore, the specific limitations of one or more embodiments of the edge computing-based intelligent monitoring and fault location method for rural power distribution provided below can be found in the limitations of the edge computing-based intelligent monitoring and fault location system for rural power distribution described above, and will not be repeated here.

[0163] In one exemplary embodiment, such as Figure 2 The diagram illustrates a process flow chart for an edge computing-based intelligent monitoring and fault location method for rural power distribution, comprising the following steps:

[0164] S1. Collect time-series data of each edge node of the power distribution network based on a unified protocol, and perform data cleaning, normalization and time alignment on the time-series data to obtain standard time-series data; wherein, the time-series data includes voltage, current, topology identifier and operating environment parameters;

[0165] S2. Based on standard time-series data, analyze the abnormal characteristics of voltage and current using a pre-set fault detection model to obtain fault characteristic data; wherein, the fault characteristic data is used to characterize the probability of each edge node failing.

[0166] S3. Based on the fault characteristic data and topology identifiers, calculate the probability distribution of the fault location based on the preset fault location model to obtain preliminary fault location data; wherein, the preliminary fault location data is used to characterize the distribution of possible fault locations;

[0167] S4. Based on the preliminary fault location data and operating environment parameters, correct the fault location results to obtain the fault location results; wherein, the fault location results are used to indicate the location area of ​​the fault.

[0168] Furthermore, based on standard time-series data and a pre-defined fault detection model, abnormal voltage and current characteristics are analyzed to obtain fault characteristic data, including:

[0169] S11. Based on the mathematical morphology gradient operator, feature extraction is performed on the voltage and current sequences in the standard time series data to capture the transient change information of the waveform and obtain the gradient feature vector.

[0170] S12. Calculate the energy entropy of the gradient feature vector within a preset time window to obtain the entropy feature value; whereby the energy entropy is used to quantify the complexity and abrupt change intensity of the signal.

[0171] S13. Input the gradient feature vector and entropy feature value into the preset lightweight random forest model for classification reasoning, and output the initial fault score of each node.

[0172] S14. Normalize the initial fault score using the Sigmoid activation function to obtain fault feature data.

[0173] Furthermore, based on the fault characteristic data and topology identifiers, the probability distribution of the fault location is calculated using a pre-defined fault location model to obtain preliminary fault location data, including:

[0174] S21. Construct the adjacency matrix of the power distribution network based on topology identifiers;

[0175] S22. Construct the fault state vector for each node based on the fault characteristic data;

[0176] S23. Based on the adjacency matrix and the fault state vector, calculate the fault propagation matrix; whereby the fault propagation matrix is ​​used to characterize the propagation intensity of fault features in the topological network; the expression for the fault propagation matrix is:

[0177]

[0178] in, This is the fault propagation matrix. It is an adjacency matrix. The Hadamard product represents the matrix multiplication, used for element-wise multiplication. For the fault state vector, It is a vector of all 1s. express transpose, This represents the absolute value operation. To prevent extremely small positive numbers with a denominator of zero;

[0179] S24. Perform weighted summation and normalization on the row vectors of the fault propagation matrix to generate the fault probability distribution of each line section and obtain preliminary fault location data.

[0180] Furthermore, based on the preliminary fault location data and operating environment parameters, the fault location results are corrected to obtain the fault location results, including:

[0181] S31. Extract the ambient temperature and ambient humidity from the operating environment parameters, and query the environmental risk coefficient corresponding to the current environmental conditions according to the preset historical fault statistics table.

[0182] S32. Construct a correction factor matrix based on environmental risk coefficients;

[0183] S33. Perform a dot product operation on the probability distribution and correction factor matrix in the preliminary fault location data to obtain the weighted corrected fault probability.

[0184] S34. Select the region with the largest value among the weighted and corrected fault probabilities as the final fault point and output the fault location result.

[0185] Furthermore, the row vectors of the fault propagation matrix are weighted, summed, and normalized to generate the fault probability distribution for each line section, thus obtaining preliminary fault location data, including:

[0186] S41. Normalize each row vector of the fault propagation matrix using the L1 norm to obtain the node fault probability value of each node.

[0187] S42. Based on the topological connection relationship of the power distribution network, calculate the fault probability of each line segment to obtain the initial probability of the line segment.

[0188] S43. The initial probability of the line section is filtered by a preset dynamic truncation threshold to remove probability sections below the preset threshold, thus obtaining a candidate fault set.

[0189] S44. The segments in the candidate fault set are aggregated using a density-based spatial clustering algorithm, and the cluster with the highest probability density is output as the preliminary fault location data.

[0190] Furthermore, a correction factor matrix is ​​constructed based on the environmental risk coefficient, including:

[0191] S51. Obtain the geographical coordinates of each edge node in the power distribution network and construct a geographical coordinate matrix;

[0192] S52. Calculate the spatial distance between nodes using the Euclidean distance formula on the geographic coordinate matrix to obtain the spatial distance matrix;

[0193] S53. Calculate the Gaussian radial basis function values ​​based on the spatial distance matrix as spatial weights, and generate a node comprehensive correction factor in combination with the environmental risk coefficient;

[0194] The expression for the comprehensive correction factor is as follows:

[0195]

[0196] In the formula, Represents a node The comprehensive correction factor, Represents a node Environmental risk coefficient and Representing nodes respectively and nodes The geographic coordinate vector, Represents a node The set of neighboring nodes, The bandwidth parameter represents the Gaussian kernel and is used to control the spatial influence range.

[0197] S54. Construct a diagonal matrix by using the node synthesis correction factor as the diagonal element to obtain the correction factor matrix.

[0198] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0199] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement a rural power distribution intelligent monitoring and fault location system based on edge computing as described above.

[0200] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0201] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0202] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A smart monitoring and fault location system for rural power distribution based on edge computing, characterized in that, The system includes: The data acquisition and preprocessing module 11 is used to acquire time-series data of each edge node of the power distribution network based on a unified protocol, and to perform data cleaning, normalization and time-series alignment on the time-series data to obtain standard time-series data; wherein, the time-series data includes voltage, current, topology identifier and operating environment parameters; The fault feature analysis module 12 is used to analyze the abnormal characteristics of voltage and current based on the standard time series data and a preset fault detection model to obtain fault feature data; wherein, the fault feature data is used to characterize the probability of each edge node failing. The preliminary fault location module 13 is used to calculate the probability distribution of the fault location based on the fault feature data and the topology identifier, and obtain preliminary fault location data; wherein, the preliminary fault location data is used to characterize the distribution of possible fault locations. The fault location correction module 14 is used to correct the fault location result based on the preliminary fault location data and the operating environment parameters to obtain the fault location result; wherein, the fault location result is used to indicate the location area of ​​the fault.

2. The system according to claim 1, characterized in that, The fault feature analysis module 12 includes: The feature extraction subunit 121 is used to extract features from the voltage and current sequences in the standard time series data based on the mathematical morphology gradient operator, capture the transient change information of the waveform, and obtain the gradient feature vector. The energy entropy calculation subunit 122 is used to calculate the energy entropy of the gradient feature vector within a preset time window to obtain the entropy feature value; wherein, the energy entropy is used to quantify the complexity and abrupt change intensity of the signal; The initial fault scoring subunit 123 is used to input the gradient feature vector and the entropy feature value into a preset lightweight random forest model for classification reasoning and output the initial fault score of each node. The fault feature subunit 124 is used to normalize the initial fault score using the Sigmoid activation function to obtain the fault feature data.

3. The system according to claim 1, characterized in that, The preliminary fault location module 13 includes: The adjacency matrix construction subunit 131 is used to construct the adjacency matrix of the power distribution network based on the topology identifier; The fault state vector construction subunit 132 is used to construct the fault state vector of each node based on the fault feature data. The fault propagation matrix calculation subunit 133 is used to calculate the fault propagation matrix based on the adjacency matrix and the fault state vector; wherein, the fault propagation matrix is ​​used to characterize the propagation intensity of fault features in the topological network; the expression of the fault propagation matrix is: in, This is the fault propagation matrix. It is an adjacency matrix. The Hadamard product represents the matrix multiplication, used for element-wise multiplication. For the fault state vector, It is a vector of all 1s. express transpose, This represents the absolute value operation. To prevent extremely small positive numbers with a denominator of zero; The preliminary fault location subunit 134 is used to perform weighted summation and normalization on the row vectors of the fault propagation matrix to generate the fault probability distribution of each line section and obtain the preliminary fault location data.

4. The system according to claim 1, characterized in that, The fault location and correction module 14 includes: The environmental risk coefficient subunit 141 is used to extract the ambient temperature and ambient humidity from the operating environment parameters, and query the environmental risk coefficient corresponding to the current environmental conditions according to the preset historical fault statistics table. Correction factor construction subunit 142 is used to construct a correction factor matrix based on the environmental risk coefficient; The fault probability correction subunit 143 is used to perform a dot product operation on the probability distribution in the preliminary fault location data and the correction factor matrix to obtain the weighted corrected fault probability. The fault location determination subunit 144 is used to select the region with the largest value in the weighted and corrected fault probability as the final fault point and output the fault location result.

5. The system according to claim 3, characterized in that, The preliminary fault location subunit 134 includes: The normalization subunit 1341 is used to normalize each row vector of the fault propagation matrix using the L1 norm to obtain the node fault probability value of each node. The line segment fault probability subunit 1342 is used to calculate the fault probability of each line segment based on the topological connection relationship of the power distribution network, and obtain the initial probability of the line segment. The candidate fault screening subunit 1343 is used to filter the initial probability of the line segment through a preset dynamic truncation threshold, remove probability segments below the preset threshold, and obtain a candidate fault set. The preliminary fault location subunit 1344 is used to aggregate the segments in the candidate fault set using a density-based spatial clustering algorithm, and output the cluster with the highest probability density as the preliminary fault location data.

6. The system according to claim 4, characterized in that, The correction factor construction subunit 142 includes: Geographic coordinate matrix construction sub-unit 1421 is used to obtain the geographic coordinates of each edge node in the power distribution network and construct the geographic coordinate matrix; The spatial distance matrix calculation subunit 1422 is used to calculate the spatial distance between nodes using the Euclidean distance formula on the geographic coordinate matrix to obtain the spatial distance matrix. The comprehensive correction factor calculation subunit 1423 is used to calculate the Gaussian radial basis function value as spatial weight based on the spatial distance matrix, and generate the node comprehensive correction factor in combination with the environmental risk coefficient; The expression for the comprehensive correction factor is as follows: In the formula, Represents a node The comprehensive correction factor, Represents a node Environmental risk coefficient and Representing nodes respectively and nodes The geographic coordinate vector, Represents a node The set of neighboring nodes, The bandwidth parameter represents the Gaussian kernel and is used to control the spatial influence range. The correction factor matrix construction subunit 1424 is used to construct a diagonal matrix by using the node comprehensive correction factor as diagonal elements to obtain the correction factor matrix.

7. A method for intelligent monitoring and fault location of rural power distribution networks based on edge computing, characterized in that, The method includes: S1. Collect time-series data of each edge node of the power distribution network based on a unified protocol, and perform data cleaning, normalization and time alignment on the time-series data to obtain standard time-series data; wherein, the time-series data includes voltage, current, topology identifier and operating environment parameters; S2. Based on the standard time series data, analyze the abnormal characteristics of voltage and current according to the preset fault detection model to obtain fault characteristic data; wherein, the fault characteristic data is used to characterize the probability of each edge node failing. S3. Based on the fault feature data and the topology identifier, calculate the probability distribution of the fault location according to the preset fault location model to obtain preliminary fault location data; wherein, the preliminary fault location data is used to characterize the distribution of possible fault locations; S4. Based on the preliminary fault location data and the operating environment parameters, correct the fault location results to obtain the fault location results; wherein, the fault location results are used to indicate the location area of ​​the fault.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the system comprising any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the system comprising any one of claims 1 to 6.

Images