Power transmission line monitoring method and system based on ad hoc network

Abstract

The invention discloses a power transmission line monitoring method and system based on an ad hoc network. The method comprises the following steps: collecting monitoring data of each sensor of a power transmission line; wherein each sensor forms a network topology according to a preset ad hoc network protocol; processing the monitoring data according to a preset time sequence feature extraction algorithm to obtain a first time sequence feature; processing the monitoring data according to a preset neural network model to obtain a first spatial feature; according to a preset multi-modal fusion model, fusing the first time sequence feature and the first spatial feature to obtain a fused feature; and processing the fusion features according to a preset fault detection model to obtain a fault detection result of the power transmission line, and executing fault recovery according to the fault detection result. According to the invention, data processing and fault detection are carried out through an efficient ad hoc network technology and a deep learning model, and real-time and accurate detection of the high-voltage transmission line fault is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of transmission line monitoring, and relates to a transmission line monitoring method and system based on an ad hoc network. Background Art

[0002] With the increasing complexity of modern power systems and the expansion of high-voltage transmission lines, ensuring their safety and stability has become the core issue in the operation of power systems. As the main backbone of power transmission, the operating state of high-voltage transmission lines is directly related to the reliability of power supply.

[0003] However, most traditional high-voltage transmission line monitoring methods rely on manual inspections and traditional sensor devices, which have problems such as limited monitoring range, poor real-time performance, and high labor costs. In addition, the network topologies in existing technologies are relatively fixed and lack adaptability, and cannot restore communication in a timely manner when network nodes fail, affecting the stability and reliability of the monitoring system. At the same time, existing fault detection methods rely on single features, resulting in low accuracy of fault identification and inability to effectively cope with complex environmental changes and equipment failures. Summary of the Invention

[0004] In view of the deficiencies of the prior art, the present application provides a transmission line monitoring method and system based on an ad hoc network, which realizes real-time and accurate detection of high-voltage transmission line faults through efficient ad hoc network technology and deep learning models for data processing and fault detection.

[0005] To achieve the above object, in the first aspect, the present invention provides a transmission line monitoring method based on an ad hoc network, including:

[0006] Collecting monitoring data of each sensor of the transmission line; wherein each of the sensors forms a network topology according to a preset ad hoc network protocol;

[0007] Processing the monitoring data according to a preset time series feature extraction algorithm to obtain a first time series feature; processing the monitoring data according to a preset neural network model to obtain a first spatial feature;

[0008] Fusing the first time series feature and the first spatial feature according to a preset multimodal fusion model to obtain a fusion feature;

[0009] Processing the fusion feature according to a preset fault detection model to obtain a fault detection result of the transmission line;

[0010] Performing fault repair according to the fault detection result.

[0011] Compared with the prior art, the embodiments of the present application have the following beneficial effects: By constructing a network topology through an ad-hoc network protocol, it ensures that each sensor can stably and efficiently collect monitoring data of the transmission line, and can quickly restore the communication path when a node fails, ensuring the continuity and integrity of data transmission; Using a preset time series feature extraction algorithm and a neural network model to process the monitoring data respectively, accurate first time series features and first spatial features are obtained, enhancing the understanding of the time and space dimensions of the fault mode; The multi-modal fusion model further integrates these two features, improving the comprehensiveness and accuracy of feature representation; Finally, through the fault detection model, efficient identification and precise positioning of complex fault modes are achieved, significantly improving the speed and accuracy of fault detection; Repair measures are executed according to the fault detection results, ensuring the safe and stable operation of the high-voltage transmission line.

[0012] In some embodiments of the first aspect of the present application, the sensors form a network topology according to a preset ad-hoc network protocol, including:

[0013] Initialize each sensor as each node of the network topology;

[0014] Obtain the broadcast response signals of each neighbor node of each node;

[0015] According to the signal strength and response time of each broadcast response signal, calculate the signal quality between each node and each neighbor node respectively, and connect each node and its neighbor node corresponding to the highest signal quality according to each signal quality to form a first network topology;

[0016] According to the signal strength, response time, node load, change metric value of the network topology and a preset Q-learning algorithm, adjust and update the first network topology to obtain the network topology.

[0017] Compared with the prior art, the above embodiments have the following beneficial effects: Based on the dynamic routing selection mechanism of the Q-learning algorithm, the transmission line system can optimize the routing path in real time, ensuring automatic reconstruction of the optimal path when a node fails or the communication quality deteriorates, guaranteeing the stability and timeliness of monitoring data transmission, and improving the reliability of fault detection.

[0018] In some embodiments of the first aspect of the present application, the sensors form a network topology according to a preset ad-hoc network protocol, further including:

[0019] Obtain the heartbeat packet data of each node according to a preset heartbeat packet mechanism;

[0020] Calculate the communication status between each node according to each heartbeat packet data;

[0021] Adjust the network topology according to the communication state and a preset dynamic routing algorithm.

[0022] Compared with the prior art, the above embodiments have the following beneficial effects: By introducing a heartbeat packet mechanism to continuously monitor the network state, the system can promptly detect node failures or communication anomalies, and quickly adjust the network structure through a dynamic routing algorithm, ensuring the integrity and real-time nature of the monitoring data, reducing data loss caused by node failures, and enhancing the effectiveness of fault detection.

[0023] In some embodiments of the first aspect of the present application, collecting the monitoring data of each sensor on the transmission line includes:

[0024] Collecting the original monitoring data of each sensor on the transmission line;

[0025] Processing the original monitoring data according to a pre-budgeted preprocessing algorithm to obtain monitoring data; wherein, the preprocessing algorithm includes any one or more combinations of the following: formatting processing, filtering and denoising processing, missing value filling processing, and standardization processing.

[0026] Compared with the prior art, the above embodiments have the following beneficial effects: By comprehensively preprocessing the original data, including steps such as removing noise, filling missing values, and standardization, the quality of the monitoring data and the accuracy of subsequent analysis are improved, providing high-quality basic data support for fault detection.

[0027] In some embodiments of the first aspect of the present application, the missing value filling processing includes:

[0028] Processing the original monitoring data according to a preset weighted K-nearest neighbor interpolation algorithm to obtain monitoring data; the algorithm is as follows:

[0029]

[0030] Where x i represents the data point to be interpolated, x j represents the known data points in the original monitoring data, d(x i , x j ) represents the Euclidean distance or Manhattan distance between the point to be interpolated and the known point, N k represents the set of K nearest neighbor points with the smallest distance from the point to be interpolated x i , w k is the weight of the k-th nearest neighbor point, Qual k is the quality value of the k-th nearest neighbor point, α represents the weighting factor, represents the interpolated data value.

[0031] Compared with the prior art, the above embodiments have the following beneficial effects: By using an improved weighted K-Nearest Neighbor (KNN) interpolation algorithm to fill in missing values, not only the distance weights between data points are considered, but also a dynamic weighting mechanism for sensor quality and data correlation is introduced, improving the accuracy and reliability of the interpolation results, ensuring the integrity and consistency of the monitoring data, and providing a more solid foundation for subsequent fault detection.

[0032] In some embodiments of the first aspect of the present application, processing the monitoring data according to a preset time series feature extraction algorithm to obtain a first time series feature includes:

[0033] Extracting the time domain features of the monitoring data according to a preset sliding window technique; wherein, the time domain features include any one or more combinations of the following: mean, variance, maximum value, minimum value, slope, and abnormal spikes;

[0034] Processing the monitoring data according to a preset fast Fourier transform algorithm to obtain frequency domain features;

[0035] Summarizing the time domain features and frequency domain features to obtain a first time series feature.

[0036] Compared with the prior art, the above embodiments have the following beneficial effects: By combining the sliding window technique and the fast Fourier transform, the short-term trends, periodic fluctuations, and sudden abnormal spikes in the time series data are effectively captured, improving the extraction accuracy of the time series features, helping to more accurately identify potential fault patterns, and enhancing the sensitivity of fault detection.

[0037] In some embodiments of the first aspect of the present application, processing the monitoring data according to a preset neural network model to obtain a first spatial feature includes:

[0038] Using a preset convolutional neural network to process the spatial distribution information in the monitoring data to obtain a local first spatial feature map;

[0039] Processing the local first spatial feature map and the adjacency matrix between each sensor node according to a preset graph convolutional network to obtain a first spatial feature.

[0040] Compared with the prior art, the above embodiments have the following beneficial effects: By processing the spatial distribution information through a convolutional neural network and a graph convolutional network, the physical position relationship and interaction between sensors are fully considered, enhancing the expression ability of the spatial features, helping to more accurately locate the fault source, and improving the spatial resolution of fault detection.

[0041] In some embodiments of the first aspect of the present application, fusing the first temporal feature and the first spatial feature according to a preset multimodal fusion model to obtain a fused feature, including:

[0042] The fusion algorithm of the multimodal fusion model is as follows:

[0043] Xfusion = α(t)·X seq +β(t)·X sp ; where Xfusion represents the fused feature, X seq represents the first temporal feature, X sp represents the first spatial feature, α(t) and β(t) are the weighting coefficients of the first temporal feature and the first spatial feature respectively, and t is the time step;

[0044] Among them, the adjustment algorithm of the weighting coefficient is as follows:

[0045] Among them, f seq (X seq ,t) and f sp (X sp ,t) respectively represent the temporal feature importance evaluation function and the spatial feature importance evaluation function, λ1 and λ2 are adjustment terms, ∥∥X seq (t)∥∥ 2 and ∥∥X sp (t)∥∥ 2 respectively represent the L2 norms of the first temporal feature and the first spatial feature.

[0046] The gradient update of the weighting coefficient is as follows:

[0047] where η is the learning rate, and respectively represent the gradients of the loss function L with respect to the weighting coefficients α and β.

[0048] Compared with the prior art, the above embodiments have the following beneficial effects: By organically combining the temporal feature and the spatial feature through a weighted fusion strategy, the model can flexibly adjust its learning strategy in different situations, improving the accuracy and robustness of fault detection; Introducing dynamic adjustment of the weighting coefficient further enhances the adaptability of the model to the importance of different features, improving the overall performance, especially for the recognition of complex fault patterns.

[0049] In some embodiments of the first aspect of the present application, the fault detection model includes: a multi-layer perceptron, a graph neural network, a long short-term memory network, a fusion layer, a fully connected layer, and a classifier;

[0050] Processing the fusion features according to a preset fault detection model to obtain a fault detection result of the transmission line, including:

[0051] Processing the fusion features according to the multi-layer perceptron to obtain a fault prediction value;

[0052] Processing the fault prediction value and the first spatial feature according to the graph neural network to obtain a second spatial feature between each node;

[0053] Processing the second spatial feature and the first temporal feature according to the long short-term memory network to obtain a second temporal feature of each node;

[0054] Fusing the second spatial feature and the second temporal feature according to the fusion layer to obtain a comprehensive feature representation;

[0055] Processing the comprehensive feature representation according to the fully connected layer and the classifier, and outputting the fault detection result of each node.

[0056] Compared with the prior art, the above embodiments have the following beneficial effects: The multi-layer perceptron (MLP) performs a non-linear transformation on the preliminarily fused spatio-temporal features, extracts potential fault patterns, and generates a preliminary fault prediction value, providing a basis for the subsequent in-depth analysis of spatial and temporal features; The graph neural network (GNN) is used to further process the fault prediction value output by the MLP and the first spatial feature, capture the complex spatial dependence relationship between sensor nodes, and through graph convolution operations, not only enhances the expression ability of spatial features, but also reveals the implicit associations between nodes, forming a more refined second spatial feature; The long short-term memory network (LSTM) is applied to combine the second spatial feature and the original first temporal feature to deeply explore the law of data change over time, effectively capture the dependencies in the long time series, thereby identifying more complex fault patterns, not only retaining the important information of the original temporal features, but also integrating the spatially enhanced features through deep learning, making the temporal features more rich and representative; In the fusion layer, the second spatial feature and the second temporal feature are weighted and integrated to form a comprehensive and detailed comprehensive feature representation, which contains information in both spatial and temporal dimensions and has been optimized by multiple rounds of deep learning models, greatly improving the recognition and discrimination of features; Finally, the fault detection result of each node is output through the fully connected layer and the classifier, and its severity level is evaluated, realizing the precise positioning and rapid response to faults.

[0057] In a second aspect, the present invention also provides a transmission line monitoring system based on an ad hoc network, including: a collection module, a first feature extraction module, a fusion module, a detection module, and a repair module;

[0058] Among them, the acquisition module is used to acquire the monitoring data of each sensor on the transmission line; each of the sensors forms a network topology according to a preset self-organizing network protocol;

[0059] The first feature extraction module is used to process the monitoring data according to a preset time-series feature extraction algorithm to obtain the first time-series feature; and process the monitoring data according to a preset neural network model to obtain the first spatial feature;

[0060] The fusion module is used to fuse the first time-series feature and the first spatial feature according to a preset multi-modal fusion model to obtain a fusion feature;

[0061] The detection module is used to process the fusion feature according to a preset fault detection model to obtain the fault detection result of the transmission line;

[0062] The repair module is used to perform fault repair according to the fault detection result.

[0063] Compared with the prior art, the above embodiments of the present application have the following beneficial effects: By constructing a network topology through the self-organizing network protocol, it ensures that each sensor can collect the monitoring data of the transmission line stably and efficiently, and can quickly restore the communication path when a node fails, ensuring the continuity and integrity of data transmission; Using the preset time-series feature extraction algorithm and neural network model to process the monitoring data respectively to obtain accurate first time-series features and first spatial features, enhancing the understanding of the time and space dimensions of the fault mode; The multi-modal fusion model further integrates these two features, improving the comprehensiveness and accuracy of feature representation; Finally, through the fault detection model, it realizes the efficient identification and precise positioning of complex fault modes, significantly improving the speed and accuracy of fault detection; Performing repair measures according to the fault detection result ensures the safe and stable operation of the high-voltage transmission line. Description of the Drawings

[0064] Figure 1 : It is a schematic flow chart of a transmission line monitoring method based on self-organizing network provided in some embodiments of the present invention.

[0065] Figure 2 : It is a schematic structural diagram of a transmission line monitoring system based on self-organizing network provided in some embodiments of the present invention.

[0066] Figure 3 : It is a comparison diagram of experimental effects provided in some embodiments of the present invention. Detailed Embodiments

[0067] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0068] Embodiment 1:

[0069] Please refer to Figure 1 , a power transmission line monitoring method based on an ad hoc network provided by an embodiment of the present invention, including steps S1 to S5:

[0070] Step S1: Collect the monitoring data of each sensor on the power transmission line.

[0071] Specifically, the monitoring data includes environmental data, electrical data, and mechanical data. Among them, the environmental data includes temperature, humidity, and wind speed; the electrical data includes current, voltage, and power factor; the mechanical data includes vibration, mechanical stress, and bending angle.

[0072] Among them, each of the sensors forms a network topology according to a preset ad hoc network protocol. Specifically, it can be implemented through the following preferred embodiments, including steps S11 - S17, as follows:

[0073] S11: Initialize each sensor as each node of the network topology.

[0074] S12: Obtain the broadcast response signals of each neighbor node of each node.

[0075] S13: Calculate the signal quality between each node and each neighbor node respectively according to the signal strength and response time of each broadcast response signal, and connect each node and its neighbor node corresponding to the highest signal quality according to each signal quality to form a first network topology.

[0076] S14: Adjust and update the first network topology according to the signal strength, response time, node load, change metric value of the network topology, and a preset Q - learning algorithm to obtain the network topology.

[0077] In specific implementation, the initialization step can be implemented using an ad hoc network protocol (such as Zigbee or LoRaWAN). Then, each sensor node searches for surrounding connectable nodes through a broadcast signal. For example, node N i sends a broadcast signal B i to search for surrounding connectable nodes. Node N j (neighbor node) after receiving the broadcast signal, responds through response packet R jSend its signal strength (RSSI) and response time (RTT) back to node N i , node N i will evaluate the connection signal quality C of each neighbor node according to RSSI and RTT ij , calculated by the following formula: where w1 and w2 represent weight factors. After each node calculates the signal quality between neighbor nodes, it selects the optimal neighbor node for connection to form the first network topology; in addition, during data transmission, each node continuously monitors the connection signal quality, and the monitoring algorithm is as follows: where Load j is the communication load of node N j , ΔTopology is a measure of the change in the current network topology, and w3 and w4 represent weight factors; when the connection signal quality changes, each node dynamically adjusts the network topology according to the Q-learning algorithm, where the Q-learning algorithm includes state monitoring, action execution, reward function, and Q-value update, and the state monitoring is as follows: s i =(RSSI i , RTT i , Load i , ΔTopology i ), where s i represents the state of node N i , action execution: the action a i of node N i means selecting which neighbor node to establish a connection with and transmitting data; reward function: node N i evaluates the quality of the transmission path through the following reward function R(s i , a i ):

[0078]

[0079] Q-value update: The Q-learning algorithm dynamically adjusts the path selection through the following update rule based on immediate reward and future reward: where α represents the learning rate and γ represents the discount factor.

[0080] In this preferred embodiment, steps S11-S14 are based on the dynamic routing selection mechanism of the Q-learning algorithm, enabling the power transmission line system to optimize the routing path in real time, ensuring automatic reconstruction of the optimal path when a node fails or the communication quality deteriorates, guaranteeing the stability and timeliness of the monitoring data transmission, and improving the reliability of fault detection.

[0081] S15: Obtain the heartbeat packet data of each node according to the preset heartbeat packet mechanism.

[0082] S16: Calculate the communication status between each node based on the heartbeat packet data of each node.

[0083] S17: Adjust the network topology according to the communication status and a preset dynamic routing algorithm.

[0084] In specific implementation, nodes continuously monitor the communication status with neighbor nodes through periodic heartbeat packets and a health check mechanism. Once the heartbeat packet of a certain node times out or the quality of the received connection signal is lower than a set threshold, it is determined that the node is a failed node. At this time, the neighbor nodes of the failed node will re-evaluate the currently available communication paths and use the dynamic routing algorithm to adjust or switch to new neighbor nodes; or when a certain node is overloaded, node N i will reselect a communication path according to the load balancing strategy to avoid network bottlenecks: where Threshold represents the load threshold and ReassignData represents reassigning the communication path;

[0085] Furthermore, in this preferred embodiment, steps S15 - S17 introduce a heartbeat packet mechanism to continuously monitor the network status, enabling the system to promptly sense node failures or communication anomalies, and quickly adjust the network structure through the dynamic routing algorithm, ensuring the integrity and real-time nature of the monitoring data, reducing data loss caused by node failures, and enhancing the effectiveness of fault detection.

[0086] Furthermore, the monitoring data collection of each sensor on the transmission line in step S1 can be achieved through the following preferred implementation methods, including steps S18 - S19, specifically as follows:

[0087] S18: Collect the original monitoring data of each sensor on the transmission line;

[0088] S19: Process the original monitoring data according to a pre-budgeted preprocessing algorithm to obtain the monitoring data; where the preprocessing algorithm includes any one or more combinations of the following: formatting processing, filtering and denoising processing, missing value supplementation processing, and standardization processing.

[0089] In specific implementation, when performing formatting processing, a time synchronization algorithm based on timestamp alignment can be used to ensure that data from different sensors can be merged and compared according to a unified time scale. By normalizing and aligning the timestamps of the collected data, the problem of data mismatch caused by different sensor sampling frequencies can be avoided, improving the accuracy and efficiency of data fusion; specifically, during the synchronization process, if there is a deviation in the sampling time of the sensor, linear interpolation is used to align the data in terms of time to ensure the time synchronization of the data;

[0090] In addition, when performing filtering and denoising processing, the Extended Kalman Filter (EKF) algorithm can be preferentially selected, as follows:

[0091] Construct a non-linear system and measurement model, which includes a system state equation and a measurement equation;

[0092] The system state equation is expressed as: x k = f(x k-1 , u k ) + w k ;

[0093] Among them, x k represents the state vector of the system, and f(x k-1 , u k ) is the system state transition function, which describes how the state x k-1 at the previous moment transfers to the state at the current moment through the control input u k (such as load changes, environmental factors, etc.), is the process noise, which reflects the unpredictable factors inside the system, such as the accuracy problem of sensors, fluctuations in the power system, etc., and Q represents the covariance matrix of the process noise.

[0094] The measurement equation is expressed as: z k = h(x k ) + v k ; Among them, z k represents the observed data of the sensor, such as the measured value of voltage or current, which contains the real-time data obtained by the sensor, and h(x k ) is the measurement function, which converts the state vector x k of the system into the observed quantity z k ; is the measurement noise, which reflects the error that may be introduced by the sensor during the measurement process, and R is the covariance matrix of the measurement noise.

[0095] The Extended Kalman Filter includes two stages: prediction and update. Among them, the prediction stage includes:

[0096] First, predict the state at the current moment using the state estimate and control input at the previous moment:

[0097] Among them, is the predicted value of the current state, which is calculated based on the estimated value at the previous moment and the control input u k .

[0098] Predict the covariance matrix: Among them, is the predicted covariance matrix at the current moment, which depends on the covariance matrix P at the previous momentk-1 and the state transition matrix F of the system k-1 and the transpose matrix where is the Jacobian matrix of the state transition function f with respect to the state, which is used to linearize the non-linear system. This matrix reflects how the system state affects each other among different dimensions.

[0099] The update phase includes: after observing the measurement value, updating the state estimate of the system by combining the predicted value and the actual measurement value, and calculating the Kalman gain: where K k is the Kalman gain represents the measurement matrix, that is, the Jacobian matrix of the measurement function with respect to the state.

[0100] Update the state estimate: where is the updated state estimate, which is corrected based on the predicted value and the measurement residual to make corrections.

[0101] Update the covariance matrix: where P k is the updated covariance matrix.

[0102] By introducing a mechanism for dynamically adjusting the process noise matrix Q and the measurement noise matrix R, considering the sensor signal quality and load conditions, the noise matrix is dynamically adjusted to improve the filtering accuracy. For example, if the signal quality of a certain node is poor, the corresponding measurement noise R is increased, thereby reducing the influence of the data of this node during the filtering process. In addition, according to the time-varying characteristics of the power system, the system state equation f can be dynamically adjusted according to the real-time network load and power demand, ensuring that the state transition model can more accurately reflect the operating conditions of the power grid.

[0103] In addition, when performing normalization processing, the robust Z-score normalization algorithm is preferably selected. Compared with the general normalization algorithm that uses the mean and standard deviation, the robust Z-score normalization algorithm uses the median and interquartile range of the data for normalization; this method has stronger anti-interference ability for high-noise data and can effectively avoid the normalization failure caused by extreme values or outliers. Specifically, first sort the data, calculate the median of the data, and use the interquartile range (that is, the difference between the 75th percentile and the 25th percentile of the data) instead of the standard deviation to achieve data normalization. This method is especially suitable for the sensor data of high-voltage transmission lines because these sensor data may contain extreme values or abnormal fluctuations, and the traditional Z-score algorithm often cannot handle this situation, while the robust Z-score algorithm can better ensure the robustness of data processing.

[0104] In this preferred embodiment, steps S18 - S19 comprehensively preprocess the original data, including steps such as removing noise, filling in missing values, and standardization, which improves the quality of the monitoring data and the accuracy of subsequent analysis, providing high-quality basic data support for fault detection.

[0105] Preferably, the missing value filling process can be implemented by the weighted K-nearest neighbor interpolation algorithm, and the algorithm is as follows:

[0106] d(x i ,x j )=∥∥x i -x j ∥∥ p ; N k ={x1,x2,…,x K};

[0107] where x i represents the data point to be interpolated, x j represents the known data points in the original monitoring data, d(x i ,x j ) represents the distance between the point to be interpolated and the known point, usually measured by the Euclidean distance p = 2 or the Manhattan distance p = 1, N k represents the set of K nearest neighbor points with the smallest distance to the point to be interpolated x i , w k is the weight of the k-th nearest neighbor point, Qual k is the quality value of the k-th nearest neighbor point, and α represents the weighting factor, represents the interpolated data value.

[0108] In addition, in specific implementation, according to the density and missing degree of the data, the value of K can be dynamically adjusted: if there is less data missing, a smaller value of K can be selected to improve the local interpolation accuracy; while when there is more data missing, a larger value of K is selected to improve the global interpolation stability.

[0109] In this preferred embodiment, the missing values are filled by the improved weighted K-nearest neighbor (KNN) interpolation algorithm, which not only considers the distance weights between data points but also introduces a dynamic weighting mechanism for sensor quality and data correlation, improving the accuracy and reliability of the interpolation results, ensuring the integrity and consistency of the monitoring data, and providing a more solid foundation for subsequent fault detection.

[0110] Step S2: Process the monitoring data according to a preset time series feature extraction algorithm to obtain first time series features; process the monitoring data according to a preset neural network model to obtain first spatial features.

[0111] Preferably, step S2 can be implemented through the following preferred embodiments, including steps S21 - S25, specifically as follows:

[0112] S21: Extract the time - domain features of the monitoring data according to the preset sliding window technique; wherein, the time - domain features include any one or more combinations of the following: mean, variance, maximum value, minimum value, slope, and abnormal spike.

[0113] S22: Process the monitoring data according to the preset fast Fourier transform algorithm to obtain frequency - domain features.

[0114] In specific implementation, the sliding window technique is adopted. By setting a window with a fixed length to slide on the time - series data, the data within the window will be used for time - series feature extraction. The window size is determined according to the sampling frequency of the high - voltage transmission line and the timeliness of data changes. For example, it is set to 30 data points, which can capture data changes in a short time and will not overly rely on local data, resulting in information loss. At each moment within the sliding window, the extracted features include but are not limited to the mean, variance, maximum value, minimum value, and slope of the data. These features can reflect the short - term trend of time - series data. In addition, for periodic fluctuations, the fast Fourier transform (FFT) can be used to extract frequency features. Specifically, in this application, it is not limited to using the fast Fourier algorithm to extract frequency features, and other similar algorithms such as the Fourier algorithm are also acceptable. For sudden anomalies, the abnormal spike can be extracted by comparing the difference between the maximum value within the window and the values of the front and rear windows.

[0115] S23: Summarize the time - domain features and frequency - domain features to obtain the first time - series features.

[0116] In this preferred embodiment, steps S21 - S23 combine the sliding window technique and the fast Fourier transform, effectively capturing the short - term trend, periodic fluctuations, and sudden abnormal spikes in the time - series data, improving the extraction accuracy of time - series features, helping to more accurately identify potential fault patterns, and enhancing the sensitivity of fault detection.

[0117] S24: Use the preset convolutional neural network to process the spatial distribution information in the monitoring data to obtain the local first spatial feature map.

[0118] S25: Process the local first spatial feature map and the adjacency matrix between each sensor node according to the preset graph convolutional network to obtain the first spatial feature.

[0119] In this preferred embodiment, steps S24 - S25 process the spatial distribution information through a convolutional neural network and a graph convolutional network, fully considering the physical position relationship and interaction between sensors, enhancing the expression ability of spatial features, contributing to more accurate localization of the fault source, and improving the spatial resolution of fault detection.

[0120] Step S3: According to the preset multi-modal fusion model, fuse the first temporal feature and the first spatial feature to obtain a fused feature.

[0121] Preferably, the multi-modal fusion model in step S3 can be fused through the following preferred implementation:

[0122] The fusion algorithm of the multi-modal fusion model is as follows:

[0123] Xfusion = α(t)·X seq +β(t)·X sp ; where Xfusion represents the fused feature, X seq represents the first temporal feature, X sp represents the first spatial feature, α(t) and β(t) are the weighting coefficients of the first temporal feature and the first spatial feature respectively, and t is the time step;

[0124] Specifically, during the weighting process of the temporal feature and the spatial feature, considering the relative importance of the features at different time steps t, a dynamic adjustment mechanism is used to optimize the weighting coefficients, and the algorithm is as follows:

[0125] where, f seq (X seq ,t) and f sp (X sp ,t) represent the temporal feature importance evaluation function and the spatial feature importance evaluation function respectively, λ1 and λ2 are adjustment terms, ∥∥X seq (t)∥∥ 2 and ∥∥X sp (t)∥∥ 2 represent the L2 norms of the first temporal feature and the first spatial feature respectively.

[0126] Specifically, the weights α(t) and β(t) need to be optimized through the backpropagation algorithm during the training process to ensure that the finally fused feature can maximize the accuracy in fault detection. The specific optimization process can be achieved through the loss function and the gradient descent algorithm. The loss function calculates the error based on the difference between the fusion result of the temporal feature and the spatial feature and the actual fault label, and the optimization process updates the weights through gradient descent. The specific algorithm is as follows:

[0127] where η is the learning rate, and respectively represent the gradients of the loss function L with respect to the weighting coefficients α and β.

[0128] In this preferred embodiment, step S3 organically combines the temporal features and spatial features through a weighted fusion strategy, enabling the model to flexibly adjust its learning strategy in different scenarios, improving the accuracy and robustness of fault detection; introducing dynamic adjustment of the weighting coefficients further enhances the model's adaptability to the importance of different features, improving the overall performance, especially for the recognition of complex fault patterns.

[0129] Step S4: Process the fused features according to a preset fault detection model to obtain a fault detection result of the transmission line; wherein, the fault detection model includes: a multi-layer perceptron, a graph neural network, a long short-term memory network, a fusion layer, a fully connected layer, and a classifier.

[0130] Preferably, step S4 can be implemented through the following preferred implementation manner, including steps S41 - S45, specifically as follows:

[0131] S41: Process the fused features according to the multi-layer perceptron to obtain a fault prediction value.

[0132] Specifically in implementation, input the fused feature Xfusion = [X seq , X sp into the multi-layer perceptron (MLP) for preliminary feature analysis. The MLP model can extract potential fault patterns through layer-by-layer non-linear transformation and mapping: Y MLP = MLP(Xfusion); where Y MLP is the fault prediction value output by the MLP model, which can usually be a fault probability value or a binary classification label (fault / non-fault).

[0133] In this preferred embodiment, step S41 performs non-linear transformation on the preliminarily fused spatio-temporal features through the multi-layer perceptron (MLP), extracts potential fault patterns, and generates preliminary fault prediction values, providing a basis for the subsequent in-depth analysis of spatial and temporal features.

[0134] S42: Process the fault prediction value and the first spatial feature according to the graph neural network to obtain the second spatial feature between each node.

[0135] Specifically in implementation, based on the result Y MLP output by the MLP, use the graph neural network (GNN) to model the spatial dependence relationship between sensor nodes: X GNN = [X sp , Y MLP ; Y GNN= GNN(X GNN ); where X GNN is the input of the GNN, and Y GNN is the second spatial feature output by the GNN.

[0136] In this preferred embodiment, step S42 further processes the fault prediction value and the first spatial feature output by the MLP using a graph neural network (GNN) to capture the complex spatial dependencies between sensor nodes. Through graph convolution operations, not only is the expression ability of the spatial feature enhanced, but also the implicit associations between nodes are revealed, forming a more refined second spatial feature.

[0137] S43: Process the second spatial feature and the first temporal feature according to the long short-term memory network to obtain the second temporal feature of each node.

[0138] Specifically, when implemented, LSTM is used to process temporal data and capture the laws of node data changing over time, especially suitable for long-term dependencies in long temporal data. At this stage, the second spatial feature extracted by the GNN and the first temporal feature are input into the LSTM model together. The calculation process of LSTM gradually extracts features through its gating mechanism (forget gate, input gate, output gate) to obtain the second temporal feature:

[0139] X LSTM = [X seq , Y GNN ; Y LSTM = LSTM(X LSTM ), where X LSTM represents the input of the LSTM, and Y LSTM represents the output second temporal feature.

[0140] In this preferred embodiment, step S43 applies the long short-term memory network (LSTM) to combine the second spatial feature and the original first temporal feature, deeply explores the laws of data changing over time, effectively captures the dependencies in long time series, thereby identifying more complex fault patterns. It not only retains the important information of the original temporal feature but also incorporates the spatially enhanced features through deep learning, making the temporal feature more rich and representative.

[0141] S44: Fuse the second spatial feature and the second temporal feature according to the fusion layer to obtain a comprehensive feature representation.

[0142] Specifically, when implemented, in order to enhance the role of spatio-temporal features in overall fault diagnosis, the second spatial feature extracted by the graph neural network and the second temporal feature extracted by the LSTM need to be weighted and fused:

[0143] Yfusion = α1Y LSTM + β1YGNN ; where Yfusion represents the comprehensive feature representation, and α1 and β1 represent the weight coefficients.

[0144] In this preferred embodiment, step S44 performs weighted integration on the second spatial feature and the second temporal feature in the fusion layer to form a comprehensive and detailed comprehensive feature representation. This comprehensive feature not only contains information in the spatio-temporal dimension but also has been optimized through multiple rounds of deep learning models, greatly improving the recognition and discrimination of features.

[0145] S45: Process the comprehensive feature representation according to the fully connected layer and the classifier, and output the fault detection results of each node.

[0146] In specific implementation, the processing of the fully connected layer and the classifier is as follows:

[0147] Fully connected layer: Y FC = ReLU(W FC Y fusion + b FC );

[0148] Among them, W FC and b FC are the weights and bias terms of the fully connected layer. The ReLU activation function is used for non-linear mapping, and Y FC represents the global feature representation.

[0149] Classifier: Among them, P(fault) represents the fault probability of each node output.

[0150] In this preferred embodiment, step S45 outputs the fault detection results of each node through the classifier and evaluates its severity level, achieving precise fault location and rapid response.

[0151] Step S5: Execute fault repair according to the fault detection results.

[0152] In specific implementation, executing the repair plan includes, according to the fault detection results, reconfiguring the network topology, selecting an effective node connection path, updating the routing information. After completing the network topology configuration, if a device fault is detected, device repair or switching to the standby power supply is performed to trigger the warning mechanism, sending the fault information to the operation and maintenance personnel, and providing the fault area or device status information. After completing the network topology repair and device repair, the fault isolation and repair program is executed.

[0153] In this embodiment, step S5 executes repair measures according to the fault detection results, ensuring the safe and stable operation of the high-voltage transmission line.

[0154] As an effect test of this application, 20 sensor nodes were deployed on a 50-kilometer high-voltage transmission line to implement the technical solution of this application. As shown in Figure 3 the experimental effect comparison diagram. Among them, the traditional method uses wired transmission. The sensor nodes are connected to the central data acquisition platform through the deployed wired network. After all data is transmitted to the platform through the wired network, preprocessing and fault detection are performed. The data preprocessing steps include Kalman filtering to remove external noise, and a threshold determination method is used for fault detection; all data is centrally processed by the central server to determine whether a fault has occurred; if an abnormality is found, repair and adjustment are performed through manual operations. Through the effect comparison, the effect of this application in various indicators is better than that of the traditional method.

[0155] In summary, compared with the prior art, the above embodiments of this application have the following beneficial effects: By constructing a network topology through the self-organizing network protocol, it ensures that each sensor can stably and efficiently collect the monitoring data of the transmission line, and quickly restores the communication path when a node fails, ensuring the continuity and integrity of data transmission; Using the preset time series feature extraction algorithm and neural network model to process the monitoring data respectively, accurate first time series features and first spatial features are obtained, enhancing the understanding of the time and space dimensions of the fault mode; The multi-modal fusion model further integrates these two features, improving the comprehensiveness and accuracy of feature representation; Finally, through a fault detection model including a multi-layer perceptron, a graph neural network, a long short-term memory network, a fusion layer, a fully connected layer, and a classifier, efficient identification and precise positioning of complex fault modes are realized, significantly improving the speed and accuracy of fault detection; Repair measures are executed according to the fault detection results, ensuring the safe and stable operation of the high-voltage transmission line.

[0156] Embodiment 2:

[0157] Please refer to Figure 2 , a transmission line monitoring system based on self-organizing network disclosed in an embodiment of the present invention includes: a collection module M1, a first feature extraction module M2, a fusion module M3, a detection module M4, and a repair module M5;

[0158] Among them, the collection module M1 is used to collect the monitoring data of each sensor of the transmission line; among them, each of the sensors forms a network topology according to the preset self-organizing network protocol.

[0159] The collection module M1 includes: an initialization unit, a response signal acquisition unit, a first topology construction unit, and a topology adjustment unit.

[0160] The initialization unit is used to initialize each sensor as each node of the network topology;

[0161] The response signal acquisition unit is used to acquire the broadcast response signals of each neighbor node of each node;

[0162] The first topology construction unit is configured to calculate the signal quality between each node and its neighbor nodes respectively according to the signal strength and response time of each broadcast response signal, and connect each node and its neighbor node corresponding to the highest signal quality according to each signal quality to form a first network topology;

[0163] The topology adjustment unit is configured to adjust and update the first network topology according to the signal strength, response time, node load, change metric value of the network topology, and a preset Q-learning algorithm to obtain a network topology.

[0164] In this embodiment, the acquisition module M1 is based on the dynamic routing selection mechanism of the Q-learning algorithm, enabling the power transmission line system to optimize the routing path in real time, ensuring automatic reconstruction of the optimal path when a node fails or the communication quality deteriorates, guaranteeing the stability and timeliness of the monitoring data transmission, and improving the reliability of fault detection.

[0165] Further, the acquisition module M1 further includes: a heartbeat packet acquisition unit, a communication status acquisition unit, and a network adjustment unit;

[0166] Among them, the heartbeat packet acquisition unit is configured to acquire the heartbeat packet data of each node according to a preset heartbeat packet mechanism;

[0167] The communication status acquisition unit is configured to calculate the communication status between each node according to each heartbeat packet data;

[0168] The network adjustment unit is configured to adjust the network topology according to the communication status and a preset dynamic routing algorithm.

[0169] The acquisition module M1 of this embodiment introduces a heartbeat packet mechanism to continuously monitor the network status, enabling the system to promptly perceive node failures or communication anomalies, and quickly adjust the network structure through a dynamic routing algorithm, ensuring the integrity and timeliness of the monitoring data, reducing data loss caused by node failures, and enhancing the effectiveness of fault detection.

[0170] Further, the acquisition module M1 further includes: a data acquisition unit and a preprocessing unit.

[0171] Among them, the data acquisition unit is configured to acquire the original monitoring data of each sensor of the power transmission line;

[0172] The preprocessing unit is configured to process the original monitoring data according to a pre-budgeted preprocessing algorithm to obtain monitoring data; where the preprocessing algorithm includes any one or more combinations of the following: formatting processing, filtering and denoising processing, missing value supplementation processing, and standardization processing.

[0173] In this embodiment, the acquisition module M1 enhances the quality of the monitoring data and the accuracy of subsequent analysis by comprehensively preprocessing the original data, including steps such as noise removal, missing value filling, and standardization, providing high-quality basic data support for fault detection.

[0174] Furthermore, the preprocessing unit includes: a missing value filling subunit.

[0175] The missing value filling subunit is used to process the original monitoring data according to a preset weighted K-nearest neighbor difference algorithm to obtain monitoring data; the algorithm is as follows:

[0176] d(x i ,x j )=∥∥x i -x j ∥∥ p ; N k ={x1,x2,…,x K};

[0177] where x i represents the data point to be interpolated, x j represents the known data points in the original monitoring data, d(x i ,x j ) represents the distance between the point to be interpolated and the known point, usually measured by the Euclidean distance p = 2 or the Manhattan distance p = 1, N k represents the set of K nearest neighbor points with the smallest distance to the point to be interpolated x i , w k is the weight of the kth nearest neighbor point, Qual k is the quality value of the kth nearest neighbor point, α represents the weighting factor, represents the interpolated data value.

[0178] In this embodiment, the missing value filling subunit fills the missing values through an improved weighted K-nearest neighbor (KNN) interpolation algorithm, which not only considers the distance weights between data points but also introduces a dynamic weighting mechanism for sensor quality and data correlation, improving the accuracy and reliability of the interpolation results, ensuring the integrity and consistency of the monitoring data, and providing a more solid foundation for subsequent fault detection.

[0179] The first feature extraction module M2 is used to process the monitoring data according to a preset time series feature extraction algorithm to obtain the first time series feature; and process the monitoring data according to a preset neural network model to obtain the first spatial feature.

[0180] The first feature extraction module M2 includes: a first time-domain feature extraction unit, a first frequency-domain feature extraction unit, and a first time-series feature summarization unit.

[0181] Among them, the first time-domain feature extraction unit is used to extract the time-domain features of the monitoring data according to the preset sliding window technique; among them, the time-domain features include any one or more combinations of the following: mean, variance, maximum value, minimum value, slope, and abnormal spike;

[0182] The first frequency-domain feature extraction unit is used to process the monitoring data according to the preset fast Fourier transform algorithm to obtain frequency-domain features;

[0183] The first time-series feature summarization unit is used to summarize the time-domain features and frequency-domain features to obtain first time-series features.

[0184] In this embodiment, the first feature extraction module M2 combines the sliding window technique and the fast Fourier transform to effectively capture the short-term trends, periodic fluctuations, and sudden abnormal spikes in the time-series data, improving the extraction accuracy of time-series features, helping to more accurately identify potential fault patterns, and enhancing the sensitivity of fault detection.

[0185] Furthermore, the first feature extraction module M2 further includes: a local space mapping unit and a first space feature extraction unit.

[0186] Among them, the local space mapping unit is used to process the spatial distribution information in the monitoring data by using a preset convolutional neural network to obtain a local first space feature map;

[0187] The first space feature extraction unit is used to process the local first space feature map and the adjacency matrix between each sensor node according to a preset graph convolutional network to obtain first space features.

[0188] In this embodiment, the first feature extraction module M2 processes the spatial distribution information through a convolutional neural network and a graph convolutional network, fully considering the physical position relationship and interaction between sensors, enhancing the expression ability of spatial features, helping to more accurately locate the fault source, and improving the spatial resolution of fault detection.

[0189] The fusion module M3 is used to fuse the first time-series features and the first space features according to a preset multi-modal fusion model to obtain fusion features.

[0190] Among them, the fusion algorithm of the multi-modal fusion model is as follows:

[0191] Xfusion = α(t)·X seq +β(t)·Xsp ; where Xfusion represents the fused feature, X seq represents the first temporal feature, X sp represents the first spatial feature, α(t) and β(t) are the weighting coefficients of the first temporal feature and the first spatial feature respectively, and t is the time step;

[0192] Among them, the adjustment algorithm of the weighting coefficient is as follows:

[0193] Among them, f seq (X seq ,t) and f sp (X sp ,t) represent the temporal feature importance evaluation function and the spatial feature importance evaluation function respectively, λ1 and λ2 are adjustment terms, ∥∥X seq (t)∥∥ 2 and ∥∥X sp (t)∥∥ 2 represent the L2 norms of the first temporal feature and the first spatial feature respectively.

[0194] The gradient update of the weighting coefficient is as follows:

[0195] where η is the learning rate, and represent the gradients of the loss function L with respect to the weighting coefficients α and β respectively.

[0196] In this embodiment, the fusion module M3 organically combines the temporal feature and the spatial feature through a weighted fusion strategy, enabling the model to flexibly adjust its learning strategy in different situations, improving the accuracy and robustness of fault detection; dynamically adjusting the weighting coefficient further enhances the adaptability of the model to the importance of different features, improving the overall performance, especially for the recognition of complex fault patterns.

[0197] The detection module M4 is used to process the fused feature according to a preset fault detection model to obtain a fault detection result of the transmission line; among them, the fault detection model includes: a multi-layer perceptron, a graph neural network, a long short-term memory network, a fusion layer, a fully connected layer, and a classifier.

[0198] The detection module M4 includes: a first detection unit, a second detection unit, a third detection unit, a fourth detection unit, and a fifth detection unit.

[0199] Among them, the first detection unit is used to process the fused feature according to the multi-layer perceptron to obtain a fault prediction value;

[0200] The second detection unit is configured to process the fault prediction value and the first spatial feature according to the graph neural network to obtain the second spatial feature between nodes;

[0201] The third detection unit is configured to process the second spatial feature and the first temporal feature according to the long short-term memory network to obtain the second temporal feature of each node;

[0202] The fourth detection unit is configured to fuse the second spatial feature and the second temporal feature according to the fusion layer to obtain a comprehensive feature representation;

[0203] The fifth detection unit is configured to process the comprehensive feature representation according to the fully connected layer and the classifier and output the fault detection result of each node.

[0204] In this embodiment, the detection module M4 performs a non-linear transformation on the preliminarily fused spatio-temporal features through a multi-layer perceptron (MLP) to extract potential fault patterns and generate preliminary fault prediction values, providing a basis for subsequent in-depth analysis of spatial and temporal features; the graph neural network (GNN) is used to further process the fault prediction values and the first spatial features output by the MLP to capture the complex spatial dependencies between sensor nodes. Through graph convolution operations, not only the expression ability of spatial features is enhanced, but also the implicit associations between nodes are revealed, forming more refined second spatial features; the long short-term memory network (LSTM) is applied to combine the second spatial features and the original first temporal features to deeply explore the laws of data changes over time, effectively capturing the dependencies in long time series, thereby identifying more complex fault patterns. It not only retains the important information of the original temporal features but also incorporates the spatially enhanced features through deep learning, making the temporal features more rich and representative; in the fusion layer, the second spatial features and the second temporal features are weighted and integrated to form a comprehensive and detailed comprehensive feature representation. This comprehensive feature contains both spatio-temporal dimensional information and has been optimized by multiple rounds of deep learning models, greatly improving the distinguishability and discrimination of features; finally, the fault detection result of each node is output through the fully connected layer and the classifier, and its severity level is evaluated, achieving precise fault location and rapid response.

[0205] The repair module M5 is configured to perform fault repair according to the fault detection result.

[0206] In this embodiment, the repair module M5 executes repair measures according to the fault detection result, ensuring the safe and stable operation of the high-voltage transmission line.

[0207] In summary, compared with the prior art, the above embodiments of the present application have the following beneficial effects: By constructing a network topology through the ad-hoc network protocol, it ensures that each sensor can stably and efficiently collect the monitoring data of the transmission line, and quickly restores the communication path when a node fails, guaranteeing the continuity and integrity of data transmission; Using the preset time series feature extraction algorithm and neural network model to process the monitoring data respectively, the accurate first time series feature and first spatial feature are obtained, enhancing the understanding of the time and space dimensions of the fault mode; The multi-modal fusion model further integrates these two features, improving the comprehensiveness and accuracy of feature representation; Finally, through the fault detection model including a multi-layer perceptron, a graph neural network, a long short-term memory network, a fusion layer, a fully connected layer, and a classifier, the efficient identification and precise positioning of complex fault modes are realized, significantly improving the speed and accuracy of fault detection; Repair measures are executed according to the fault detection results, ensuring the safe and stable operation of the high-voltage transmission line.

[0208] For the specific working processes of the above-described modules, reference may be made to the corresponding processes in the foregoing method embodiments, which will not be elaborated herein. The division of the modules is only a logical function division, and there may be other division methods in actual implementation. For example, multiple modules may be combined or integrated into another system.

[0209] The above specific embodiments have further elaborated on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the protection scope of the present invention. In particular, it is pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A transmission line monitoring method based on a self-organizing network, characterized in that: include: Collect monitoring data from various sensors on the transmission line; Each of the sensors forms a network topology according to a preset self-organizing network protocol; Processing the monitoring data according to a preset time series feature extraction algorithm to obtain a first time series feature; Processing the monitoring data according to a preset neural network model to obtain a first spatial feature; According to a preset multimodal fusion model, the first temporal feature and the first spatial feature are fused to obtain a fused feature; According to a preset fault detection model, the fusion features are processed to obtain a fault detection result of the transmission line; According to the fault detection result, fault repair is performed.

2. A transmission line monitoring method based on an ad hoc network as claimed in claim 1, characterized in that: Each of the sensors forms a network topology according to a preset self-organizing network protocol, including: Initialize each sensor as a node in the network topology; Obtaining broadcast response signals from each neighboring node of each node; Calculate the signal quality between each node and each neighboring node according to the signal strength and response time of each broadcast response signal, and connect each node and its corresponding neighboring node with the highest signal quality according to each signal quality to form a first network topology; According to the signal strength, response time, node load, network topology change measurement value and a preset Q learning algorithm, the first network topology is adjusted and updated to obtain a network topology.

3. A transmission line monitoring method based on ad hoc network as claimed in claim 2, characterized in that: Each of the sensors forms a network topology according to a preset self-organizing network protocol, and further includes: According to the preset heartbeat packet mechanism, obtain the heartbeat packet data of each node; Calculate the communication status between nodes based on the heartbeat packet data; The network topology is adjusted according to the communication status and a preset dynamic routing algorithm.

4. A transmission line monitoring method based on ad hoc network as claimed in claim 1, characterized in that: The monitoring data of each sensor of the power transmission line is collected, including: Collect the original monitoring data of each sensor of the transmission line; The raw monitoring data is processed according to the budgeted preprocessing algorithm to obtain monitoring data; wherein the preprocessing algorithm includes any one or more combinations of the following: formatting processing, filtering and denoising processing, missing value supplementation processing, and standardization processing.

5. A transmission line monitoring method based on ad hoc network as claimed in claim 4, characterized in that: The missing value supplementation process includes: The original monitoring data is processed according to a preset weighted K-nearest neighbor difference algorithm to obtain monitoring data; wherein the algorithm is as follows: d(x i ,x j )=∥∥x i -x j ∥∥ p ; N k ={x1,x2,…,x K }; where x i represents the data point to be interpolated, x j represents the known data points in the original monitoring data, d(x i ,x j ) represents the Euclidean distance or Manhattan distance between the interpolation point and the known point, N k Represents the interpolation point x i The set of K neighboring points with the smallest distance, w k is the weight of the kth neighboring point, Qual k is the quality value of the kth neighboring point, α represents the weighting factor, Represents the interpolated data value.

6. A transmission line monitoring method based on ad hoc network as claimed in claim 1, characterized in that: The step of processing the monitoring data according to a preset time series feature extraction algorithm to obtain a first time series feature includes: According to a preset sliding window technology, the time domain features of the monitoring data are extracted; wherein the time domain features include any one or more combinations of the following: mean, variance, maximum value, minimum value, slope and abnormal peak; Processing the monitoring data according to a preset fast Fourier transform algorithm to obtain frequency domain features; The time domain features and frequency domain features are aggregated to obtain a first time series feature.

7. A transmission line monitoring method based on ad hoc network as claimed in claim 1, characterized in that: The step of processing the monitoring data according to a preset neural network model to obtain a first spatial feature includes: Using a preset convolutional neural network, processing the spatial distribution information in the monitoring data to obtain a local first spatial feature map; According to a preset graph convolutional network, the local first spatial feature map and the adjacency matrix between each sensor node are processed to obtain the first spatial feature.

8. A power transmission line monitoring method based on ad hoc network as claimed in claim 1, characterized in that: The step of fusing the first temporal feature and the first spatial feature according to a preset multimodal fusion model to obtain a fusion feature includes: The fusion algorithm of the multimodal fusion model is as follows: Xfusion=α(t)·X seq +β(t)·X sp ; Xfusion represents the fusion feature, X seq represents the first time series feature, X sp represents the first spatial feature, α(t) and β(t) are the weighting coefficients of the first temporal feature and the first spatial feature, respectively, and t is the time step; The adjustment algorithm of the weighting coefficient is as follows: Among them, f seq (X seq ,t) and f sp (X sp , t) represent the temporal feature importance evaluation function and the spatial feature importance evaluation function respectively, λ1 and λ2 are adjustment terms, ∥∥X seq (t)∥∥ 2 and ∥∥X sp (t)∥∥ 2 Represent the L2 norm of the first temporal feature and the first spatial feature respectively; The gradient update of the weighting coefficient is as follows: Where η is the learning rate, and They represent the gradient of the loss function L with respect to the weighting coefficients α and β respectively.

9. A power transmission line monitoring method based on an ad hoc network as claimed in any one of claims 1 to 8, characterized in that: The fault detection model includes: a multi-layer perceptron, a graph neural network, a long short-term memory network, a fusion layer, a fully connected layer and a classifier; The processing of the fusion features according to the preset fault detection model to obtain the fault detection result of the transmission line includes: According to the multi-layer perceptron, the fusion feature is processed to obtain a fault prediction value; According to the graph neural network, the fault prediction value and the first spatial feature are processed to obtain a second spatial feature between each node; According to the long short-term memory network, the second spatial feature and the first temporal feature are processed to obtain a second temporal feature of each node; According to the fusion layer, the second spatial feature and the second temporal feature are fused to obtain a comprehensive feature representation; According to the fully connected layer and the classifier, the comprehensive feature representation is processed to output the fault detection result of each node.

10. A power transmission line monitoring system based on a self-organizing network, characterized in that: include: An acquisition module, a first feature extraction module, a fusion module, a detection module and a repair module; Wherein, the acquisition module is used to collect monitoring data of each sensor of the power transmission line; wherein each of the sensors forms a network topology according to a preset self-organizing network protocol; The first feature extraction module is used to process the monitoring data according to a preset time series feature extraction algorithm to obtain a first time series feature; and to process the monitoring data according to a preset neural network model to obtain a first spatial feature; The fusion module is used to fuse the first temporal feature and the first spatial feature according to a preset multimodal fusion model to obtain a fusion feature; The detection module is used to process the fusion features according to a preset fault detection model to obtain a fault detection result of the transmission line; The repair module is used to perform fault repair according to the fault detection result.

Images