Multi-dimensional sensing panoramic monitoring method and system for power transmission line

Abstract

The invention discloses a multi-dimensional sensing panoramic monitoring method and system for a power transmission line, and the method comprises the steps: obtaining the multi-dimensional sensing data of a collection point for each collection point of the power transmission line, and carrying out the preprocessing of the multi-dimensional sensing data; feature engineering processing is carried out on the preprocessed multi-dimensional perception data of all the collection points, and feature vectors of all the collection points are obtained through fusion; constructing a random forest regression model; completing power transmission line health degree definition and generating a power transmission line health degree label; selecting the feature vector of each collection point marked with the health degree in the historical power transmission line as training data, and inputting the training data into the constructed random forest regression model for model training; and inputting the currently obtained feature vectors of the collection points into the trained random forest regression model, obtaining and analyzing health degree evaluation results of the collection points, and generating health degree early warning. According to the invention, the health degree of the power transmission line can be accurately evaluated and early warning can be carried out timely, and the monitoring efficiency and safety of the power system are improved.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent monitoring of power transmission lines, and in particular to a multi-dimensional perception panoramic monitoring method and system for power transmission lines. Background Art

[0002] With the continuous development of power systems, the importance of intelligent monitoring in the power sector has become increasingly prominent. In applications such as remote monitoring of power transmission lines and preventive maintenance of power equipment, effective monitoring systems can ensure the stable operation of power systems, reduce the probability of failures, and mitigate the economic losses and social impact caused by power outages and other failures.

[0003] Traditionally, power monitoring employs a variety of methods to monitor transmission lines and related power equipment. Simple fixed monitoring equipment, such as those capable of basic image capture and temperature monitoring sensors, is installed on power towers to obtain limited line status information. Regular manual inspections are also implemented, with operations and maintenance personnel conducting on-site checks on the condition of transmission lines and equipment.

[0004] However, the existing power monitoring system has obvious defects, the information island phenomenon is serious, and there is a lack of effective data fusion mechanism between different types of monitoring equipment, which makes it impossible to realize the fusion analysis of multi-dimensional data. As a result, when conducting health warnings, the accuracy is greatly reduced due to the lack of comprehensive data support, making it difficult to quickly and accurately determine health warnings. Summary of the Invention

[0005] In order to promptly detect situations where the health of transmission lines does not meet standards and generate early warnings, solve the information island problem of existing power monitoring systems, realize multi-dimensional data fusion analysis, and improve early warning capabilities, this application provides a multi-dimensional perception panoramic monitoring method and system for transmission lines.

[0006] In a first aspect, the present application provides a multi-dimensional perception panoramic monitoring method for a transmission line, comprising: For each collection point on the transmission line, multi-dimensional perception data is obtained, including: image data of the collection point, time series data of various sensors, lidar point cloud data of the distance to obstacles around the collection point, and environmental data, and pre-processed for each collection point; Feature engineering is performed on the multi-dimensional perception data of each collection point after preprocessing, and the feature vectors of each collection point are obtained by fusion. The feature engineering process includes: using a single-modal feature extraction method to obtain image features, various sensor time series features, and lidar point cloud features; using a cross-modal interactive feature construction method to obtain cross-modal interactive features based on physical rules and cross-modal interactive features based on correlation analysis rules; and using a Gini importance screening method to complete feature screening. Construct a random forest regression model; complete the definition of transmission line health and generate a transmission line health label; select the feature vectors of each collection point marked with health in the historical transmission line as training data, input it into the constructed random forest regression model, and obtain a trained random forest regression model; The feature vectors of each collection point currently obtained are input into the trained random forest regression model to obtain the health assessment results of each collection point; the health assessment results of each collection node are compared with the preset health threshold, and a health warning is generated for the collection node whose health is less than the preset health threshold.

[0007] By adopting the above solution, feature engineering processing and fusion are performed on the multi-dimensional perception data of each collection point, which can extract effective features from multi-source heterogeneous data and improve data quality and availability. The constructed random forest regression model is used to obtain health assessment results and compare them with the preset health threshold. In this way, substandard health of transmission lines can be detected in time and early warnings can be generated, solving the information island problem of the existing power monitoring system, realizing multi-dimensional data fusion analysis, and improving early warning capabilities.

[0008] Preferably, it also includes: For each time series feature after feature engineering, use the pre-built LSTM prediction model to predict the features of the corresponding future time period; The features of the future time period predicted for each time series feature processed by feature engineering are combined with the features processed by feature engineering to obtain a new feature vector. The new feature vectors of each collection point currently obtained are respectively input into the retrained random forest regression model, and the health assessment results of each collection point are obtained to replace the original health assessment results of each collection point. Among them, the new feature vectors of each collection point marked with health in the historical transmission line are selected as incremental training data to retrain the random forest regression model; Construct an isolation forest anomaly detection model; use the historical data of each dimensional feature predicted by the LSTM prediction model as input, train the isolation forest anomaly detection model to identify the normal feature data and abnormal feature data of each dimension, and output the marked abnormal feature data of each dimension; use the trained isolation forest anomaly detection model to determine whether the dimensional features of the LSTM prediction model corresponding to the future time period have abnormal feature data tags, and combine with the deep learning algorithm to trigger the preset type of abnormal warning for the dimensional feature data with abnormal data tags.

[0009] By adopting the above solution, the pre-built LSTM prediction model is used to predict the features of future time periods based on the time series features processed by feature engineering, and the features are integrated with the original features into a new feature vector. The random forest regression model is retrained to evaluate the health of the collection points, thereby improving the accuracy and timeliness of the health assessment and reflecting the future health status of the transmission line. An isolation forest anomaly detection model is constructed to identify normal and abnormal data in the feature dimension predicted by the LSTM prediction model. The deep learning algorithm is combined to trigger early warnings for abnormal data, timely discover potential anomalies of the transmission line, and ensure the safe and stable operation of the transmission line.

[0010] Preferably, the method further includes: collecting and acquiring terrain data and equipment data at each collection point of the transmission line; Based on the environmental data, terrain data, and equipment data of each collection point in the transmission line, determine the scenario type corresponding to the current collection point, including: normal state scenario and abnormal state scenario; the abnormal state scenario includes: extreme weather scenario, complex terrain scenario, external force damage scenario, and equipment aging scenario; In order to determine that the current acquisition point is an abnormal scene, feature engineering optimization processing is performed on the pre-processed multi-dimensional perception data of the current acquisition point to replace feature engineering processing, and the feature vectors of each acquisition point are obtained by fusion; the feature engineering optimization processing includes: using a single-modal feature extraction method to obtain image features, various sensor time series features, and lidar point cloud features; using a cross-modal interactive feature construction method to obtain cross-modal interactive features constructed based on physical rules that conform to the current abnormal scene, and cross-modal interactive features constructed based on correlation analysis rules; and using a Gini importance screening method to complete feature screening; Construct multiple random forest regression models, each of which is trained using the feature vectors of each historical collection point under a scenario; obtain the corresponding random forest regression model based on the currently determined abnormal scenario, and input the currently obtained feature vectors of each collection point into the trained random forest regression model to obtain the corresponding health assessment results of each collection point.

[0011] By adopting the above scheme, terrain data and equipment data are collected at each collection point of the transmission line, and the scene type is determined by combining environmental data to identify abnormal scenarios. Feature engineering optimization processing is performed on the multi-dimensional perception data in abnormal scenarios to obtain feature vectors that conform to the physical rules of abnormal scenarios. Multiple random forest regression models are constructed for different scenarios, and the corresponding models are selected according to the abnormal scenarios to evaluate the health, thereby improving the accuracy of transmission line health assessment in abnormal scenarios.

[0012] Preferably, it also includes: Define the transmission line health evaluation based on multi-dimensional indicators, including structural health, electrical health, and environmental health; For each collection point of the transmission line, the status data of the corresponding multi-dimensional perception data collection device is obtained, and the weights of indicators of different dimensions are dynamically adjusted according to the status data of the multi-dimensional perception data collection device; including: separately counting the number of devices with fault risks according to the status data of the corresponding collection devices of the corresponding dimensional perception data of structural health, electrical health and environmental health, and dynamically adjusting the weights of indicators of different dimensions according to the proportion of the number of devices with fault risks counted in different dimensions.

[0013] By adopting the above scheme, the health of the transmission line is defined as a weighted evaluation of multi-dimensional indicators to comprehensively measure the health status; the weights of indicators of different dimensions are dynamically adjusted according to the status data of the multi-dimensional perception data acquisition device, so that the health assessment fits the actual situation, avoids the impact of the failure of some acquisition devices on the assessment accuracy, and improves the reliability and effectiveness of the assessment results.

[0014] Preferably, the random forest regression model construction includes: Initialize random forest parameters, including the number of decision trees, maximum depth, and minimum number of samples for splitting nodes; Bootstrap sampling is used to extract sample subsets with replacement from the feature vector of the original data, and multiple decision trees are constructed. When each decision tree splits, some features are randomly selected to obtain the final decision trees as training data. The random forest regression model is fitted using the training data. The input of the model is the feature vector obtained from the collection point. A single decision tree is used to predict the output. The mean of the prediction output of all single decision trees is calculated or the K-means clustering method is used to obtain the cluster mean to obtain the overall prediction output of the forest and obtain the health assessment result.

[0015] By adopting the above scheme, the training data is used to fit the random forest regression model, and the characteristic vector of the collection point is used as input. The overall prediction output of the forest is obtained by calculating the mean of the prediction output of a single decision tree or the K-means cluster mean, so as to obtain the health assessment results of each collection point of the transmission line more accurately.

[0016] Preferably, it also includes: The edge-cloud collaborative computing architecture is deployed to complete the health warning and abnormality warning of the transmission line; including: using the edge computing center to complete the feature engineering processing of multi-dimensional perception data, feature prediction of future time periods, and abnormality warning using the isolation forest detection model; using the cloud computing center to complete the use of the random forest regression model to obtain the health assessment results.

[0017] By adopting the above solution, the edge-cloud collaborative computing architecture is used to deploy health warnings and abnormality warnings for transmission lines, reducing the pressure of data transmission, improving processing efficiency and the real-time nature of abnormality warnings, and ensuring the accuracy and stability of health assessments.

[0018] Preferably, it also includes: After completing feature screening using the Gini importance screening method, VIF collinearity test is performed on the screened features to obtain the variance inflation coefficient of each screened feature, and features whose variance inflation coefficient is not greater than the preset variance inflation coefficient are retained.

[0019] By adopting the above scheme, VIF collinearity detection is performed on the screened features, and features with variance inflation coefficients not greater than the preset value are retained, avoiding the interference of collinearity between features on the model, and improving the accuracy of random forest regression model in transmission line health assessment.

[0020] In a second aspect, the present application provides a multi-dimensional perception panoramic monitoring system for power transmission lines, including: The transmission line data acquisition module is used to obtain multi-dimensional perception data for each collection point on the transmission line, including image data of the collection point, time series data of various sensors, lidar point cloud data of the distance to obstacles around the collection point, and environmental data, and perform pre-processing on each collection point; The transmission line data processing module is used to perform feature engineering processing on the pre-processed multi-dimensional sensing data of each collection point, and fuse them to obtain the feature vector of each collection point. The feature engineering processing includes: using a single-modal feature extraction method to obtain image features, various sensor time series features, and lidar point cloud features; using a cross-modal interactive feature construction method to obtain cross-modal interactive features based on physical rules and cross-modal interactive features based on correlation analysis rules; and using a Gini importance screening method to complete feature screening. The transmission line health assessment module is used to construct a random forest regression model; complete the definition of transmission line health and generate a transmission line health label; select the feature vectors of each collection point marked with health in the historical transmission line as training data, input the constructed random forest regression model, and obtain a trained random forest regression model; input the currently obtained feature vectors of each collection point into the trained random forest regression model respectively, and obtain the corresponding health assessment results of each collection point; compare the health assessment results of each collection node with the preset health threshold respectively, and generate a health warning for the collection node whose health is less than the preset health threshold.

[0021] By adopting the above scheme, a transmission line data acquisition module is designed to obtain and preprocess the multi-dimensional perception data of each collection point of the transmission line, providing a comprehensive and accurate data basis for subsequent analysis; a transmission line data processing module is designed to perform feature engineering and fusion on the preprocessed data to effectively extract data features; a transmission line health assessment module is designed to accurately assess the health of each collection point by constructing and training a random forest regression model, and to issue early warnings for collection nodes whose health is lower than the preset value, thereby realizing real-time monitoring and effective management of the transmission line.

[0022] In a third aspect, the present application provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the method as described above.

[0023] In a fourth aspect, the present application provides a computer device, which includes a memory, a processor, and a program stored and executable on the memory, and the program implements the steps of the above method when executed by the processor.

[0024] In summary, this application has the following beneficial effects: 1. Integrate multiple types of sensors to collect multi-dimensional perception data and perform pre-processing, feature engineering, and fusion. The data is then input into a built and trained random forest regression model to assess the health of transmission lines, promptly identifying potential faults and abnormalities. This solves the information silo problem of existing power monitoring systems, enables fusion analysis of multi-dimensional data, improves early warning capabilities, and ensures stable operation of the power system. 2. Utilize the LSTM prediction model to predict the future time period characteristics of time series features and fuse them with the original features, retraining the random forest regression model to obtain more accurate health assessment results. Simultaneously, construct an isolation forest anomaly detection model to identify abnormal data. Combined with a deep learning algorithm, this model triggers warnings based on abnormal data, improving the ability to predict and warn of potential abnormalities in transmission lines and ensuring their stable operation. 3. Consider abnormal scenarios, perform feature engineering optimization on abnormal scenarios, build a random forest regression model that adapts to different scenarios, and improve the accuracy of transmission line health assessment under abnormal scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 Flowchart of the multi-dimensional perception panoramic monitoring method for power transmission lines described in a specific embodiment; Figure 2 It is a structural diagram of the multi-dimensional perception panoramic monitoring system for transmission lines described in a specific embodiment. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0027] This application mainly adopts a solution of collecting multi-dimensional data of transmission lines to evaluate health and issue early warnings, which achieves the effect of breaking information silos and improving health early warning capabilities. The following is a further detailed description of this application.

[0028] like Figure 1 As shown, the embodiment of the present application discloses a multi-dimensional perception panoramic monitoring method for transmission lines, including: data collection and preprocessing, feature engineering processing, model construction and training, health assessment and early warning, etc. The steps are introduced step by step below.

[0029] S1. For each collection point of the transmission line, obtain multi-dimensional perception data of the collection point.

[0030] Specifically, several key points (such as poles and conductors) are selected in the transmission line as collection points, and data acquisition devices are installed at the collection points to collect multi-dimensional perception data of the collection points, including image data of the collection points (visible light images, infrared images, etc.), time series data of various sensors (such as: conductor temperature, ice thickness, insulator leakage current, partial discharge, tower inclination angle, etc.), lidar point cloud data of the distance between the collection points and the obstacles around them (such as: the minimum distance from the conductor to the ground, the distance from the conductor to the top of the tree), and environmental data.

[0031] Image data can be collected using visual monitoring equipment on towers, such as video surveillance equipment and image capture devices. These devices are typically installed at appropriate locations on the towers to provide a comprehensive and clear view of the transmission line and surrounding conditions. Time series data from various sensors is collected using a variety of sensors, such as conductor temperature monitoring sensors for conductor temperature data and conductor galloping status monitoring sensors for conductor galloping status data. These sensors must be securely mounted on the towers or conductors to ensure accurate data collection. LiDAR point cloud data is collected by LiDAR sensors mounted on drones or other mobile devices, providing distance information between the collection point and surrounding obstacles (such as trees and buildings). Environmental data is collected using environmental monitoring sensors, such as small integrated weather stations and icing sensors.

[0032] Preprocessing is performed on each collection point; the preprocessing process includes data cleaning, removing noise and erroneous data from the collected data, format unification, and converting data in different formats into a comparable unified format; data cleaning can use filtering algorithms, such as Gaussian filtering to remove noise, and format unification can formulate corresponding conversion rules based on the data type.

[0033] S2. Perform feature engineering on the pre-processed multi-dimensional perception data of each collection point, and fuse them to obtain the feature vector of each collection point; In order to avoid the phenomenon of information islands and accurately complete the fusion analysis of multi-dimensional data, the key lies in processing data features. This application sets up feature engineering processing technology including multi-layer data processing to achieve good fusion effects and assist in subsequent health assessments.

[0034] The feature engineering process specifically includes three layers of processing. The first layer is single-modal feature extraction, which extracts the data of the specific dimension currently collected to perform feature extraction of the current dimension; the second layer considers the physical coupling relationship and mutual correlation relationship between multiple dimensional data, thereby constructing cross-modal interactive features; the third layer is based on the feature data obtained in the first two layers, ensuring the subsequent accurate health assessment, calculating the feature importance, and selecting features with high importance; the processing content of each layer will be specifically explained below.

[0035] First, for the use of single-modal feature extraction methods to obtain image features, various sensor time series features and lidar point cloud features, the entire feature extraction can be set to include: dynamic feature extraction, static feature extraction and spatiotemporal feature extraction.

[0036] Image features are obtained from image data. Image feature extraction can use algorithms such as edge detection and feature point matching, or model algorithms such as YOLOv5 / ResNet and GLCM to extract static features such as morphological features and texture features. These include: the proportion of damaged insulator area, contamination level, broken conductors, and segmentation of tower corroded areas. For example, in the identification of broken conductors, the image is first preprocessed and a 5×5 Gaussian filter is used for noise reduction. The gradient is then calculated using the Sobel operator, and the edge is located using the Canny algorithm. Finally, the defect size is calculated. Time series image analysis is also considered to extract dynamic features such as the conductor's dancing trajectory and the ice growth rate.

[0037] For the time series characteristics of various sensors, feature extraction can extract time domain features through sliding windows, such as: temperature, mean / variance of vibration intensity, vibration kurtosis coefficient, leakage current pulse count of insulators, etc.; extract frequency domain features through FFT transformation, wavelet packet energy entropy, etc., such as: conductor galloping fundamental frequency, non-stationary characteristics of vibration signals, etc.; extract spatiotemporal correlation features through correlation analysis technology; and extract trend features of time series data through indicators such as hourly / daily change rate and mutation amplitude, such as: hourly change rate of tower inclination / ice thickness (hourly change rate = (current value - value 1 hour ago) / 1 hour), daily change rate (daily change rate = (current value - value 24 hours ago) / 24 hours), and mutation amplitude (mutation amplitude = current value - sliding window mean).

[0038] In view of the characteristics of lidar point cloud data, the lidar point cloud data needs to be preprocessed first, and the point cloud types need to be separated using a random forest classifier. Then, Poisson reconstruction is performed to generate a continuous three-dimensional model, and then features such as wire sag, ground distance, and tree obstacle height are extracted. Specifically, these include: the minimum Euclidean distance between the wire and the tree, the minimum distance from the wire to the ground, and the surface fitting of the wire's ground clearance height (catenary equation fitting).

[0039] Secondly, the cross-modal interaction feature construction method is used to obtain cross-modal interaction features constructed based on physical rules and cross-modal interaction features constructed based on correlation analysis rules.

[0040] The cross-modal interaction features constructed based on physical rules can be constructed according to the physical characteristics and operating laws of the transmission line. For example, combining the material formula, the coupling characteristics of ice thickness and ambient temperature are calculated to characterize the ice stress caused by sudden temperature changes, and the cross-modal interaction features = ice thickness Ambient temperature gradient; another example: Based on electromagnetic principles, combined with leakage current pulse count and relative humidity, a cross-modal interaction feature is constructed to characterize the accelerated insulation degradation caused by moisture and pollution = insulator leakage pulse Ambient humidity; Similarly, there is a cross-modal interaction feature that combines the wire dancing amplitude and wind speed to characterize the wind-induced dancing enhancement effect = wire dancing amplitude Ambient wind speed, etc.

[0041] The cross-modal interaction features constructed based on correlation analysis rules are constructed by analyzing the correlation between different modal data, such as: calculating the tower tilt-wind speed cross-correlation function, calculating the nonlinear correlation strength with tree distance and conductor vibration amplitude through mutual information; and Granger causality test results of ice thickness and ambient temperature and humidity.

[0042] Then, feature screening is completed using the Gini importance screening method; this includes: performing Gini importance sorting on the features obtained at each of the above-mentioned collection points, setting an importance threshold (such as 0.01), retaining features greater than the importance threshold, and eliminating other non-important features. In addition, further optimization of the important feature screening strategy can be planned, including: after completing feature screening using the Gini importance screening method, performing VIF collinearity testing on the screened features, obtaining the variance inflation coefficient of each screened feature, retaining features with a variance inflation coefficient no greater than the preset variance inflation coefficient, and finally using SHAP values ​​to explain feature contributions to avoid Gini importance bias.

[0043] Finally, the features of each collection point are fused according to the above features, such as feature splicing, to obtain the multi-dimensional feature vectors corresponding to the fusion of each collection point.

[0044] S3. Build a random forest regression model; complete the definition of transmission line health and generate transmission line health labels; train the random forest regression model, and use the trained random forest regression model to complete the health assessment results of each collection point.

[0045] First, a random forest regression model is built.

[0046] Initialize random forest parameters, including the number of decision trees, maximum depth, and minimum number of samples for splitting nodes; obtain training data.

[0047] Bootstrap sampling is used to extract sample subsets from the feature vectors of the original data with replacement, and multiple decision trees are constructed. When each decision tree splits, some features are randomly selected to obtain the final decision trees as training data. Among them, the feature vectors of the original data refer to the feature vectors after the historical collection points are standardized. When splitting the decision tree, the decision tree splitting criterion must be followed: the mean square error is minimized, and the split is randomly selected. features (reduce the correlation between trees); the split threshold t satisfies: Where N is the number of samples at the current node, 、 are the mean health index of the left and right subtree samples respectively; y is the sample label (health index marked by experts).

[0048] The training data is used to fit a random forest regression model. The model input is the feature vector obtained from the collection point, and the output is the collection point health assessment result. To further optimize the random forest regression model, this embodiment uses a collaborative decision-making process between individual trees and the forest to obtain the final collection point health assessment result. This includes: pre-predicting the output using a single decision tree, calculating the mean of the prediction output of all individual decision trees, or using K-means clustering to obtain the cluster mean to obtain the overall forest prediction output, and obtaining the health assessment result.

[0049] Assume that the i-th decision tree in the forest is , ( is a standardized feature vector); suppose the forest contains m decision trees, and the final health index H is the mean of the prediction values ​​of all trees: Alternatively, K-means clustering is used to calculate the cluster mean, eliminating the predicted values ​​of individual trees that are far away from the cluster center and retaining the remaining ,get , the final health value can be mapped to a percentage system.

[0050] Secondly, complete the definition of transmission line health and generate transmission line health labels.

[0051] Specifically, the health label of each collection point in the transmission line can be set to a percentage system. The specific health label (score) settings are shown in Table 1 below: Table 1 Health Label

[0052] In addition, to better improve the setting of health labels, it is also possible to define a multi-dimensional indicator weighted evaluation of transmission line health, including structural health, electrical health, and environmental health; the formula is: .

[0053] Accordingly, the training data is used to fit the random forest regression model. The input of the model is the feature vector obtained from the collection point, and the output is the health assessment result process of the collection point, which includes first obtaining the structural health, electrical health and environmental health, and then obtaining them through weighted calculation.

[0054] The weighting of each health indicator can be dynamically adjusted based on the status data of the multi-dimensional sensing data collection device. Specifically, the following steps are performed: obtaining the corresponding multi-dimensional sensing data collection device status data for each collection point on the transmission line, including whether there is a fault risk; counting the number of devices with fault risk corresponding to the corresponding sensing data collection devices for the structural health, electrical health, and environmental health dimensions, and dynamically adjusting the weights of the different dimensional indicators based on the proportion of the number of devices with fault risks counted in each dimension. For example, the multi-dimensional sensing data settings are pre-specified to correspond to health attributes, such as: the tower tilt angle belongs to structural health, the leakage current peak belongs to electrical health, and the distance between the conductor and trees belongs to environmental health.

[0055] Then, the constructed random forest regression model is trained using the training data.

[0056] Specifically, the historical feature vectors of each collection point on the transmission line and the health of each collection point labeled based on expert experience are obtained as training data, which are divided into training sets and test sets. The training set is pre-input into the constructed random forest regression model. The parameters of the model are optimized through multiple iterations to obtain a trained random forest regression model; the model trained with the test set is then evaluated to calculate the model's predictive performance indicators.

[0057] In addition, once the multidimensional health indicators such as structural health, electrical health and environmental health are set, the historical feature vectors of each collection point of the transmission line and the multidimensional health of each collection point labeled based on expert experience are obtained as training data during the corresponding training process to complete the training.

[0058] Finally, the trained random forest regression model is used to complete the health assessment results of each collection point.

[0059] Specifically, the currently acquired feature vectors of each collection point are input into the trained random forest regression model respectively, and the health assessment results of each collection point are correspondingly obtained, including the health value and status level.

[0060] S4. Generate a health warning based on the health assessment results of each collection point.

[0061] Specifically, the health assessment results of each collection node are compared with a preset health threshold (e.g., 60 points). Health alerts are generated for collection nodes with health scores below the preset threshold. Accordingly, collection points with poor health status are identified as fault points, generating severe health alerts and promptly notifying the collection point location for repair.

[0062] The implementation principles of this embodiment are as follows: by comprehensively collecting multi-dimensional perception data of the transmission lines, preprocessing and feature engineering the data, and fully mining the effective information in the data; using the random forest regression model to evaluate the health of the transmission lines; and generating early warnings by comparing with preset health levels to promptly discover potential problems with the transmission lines, thereby improving the health warning capability and ensuring the stable operation of the power system.

[0063] In a specific embodiment, in order to further improve the accuracy and timeliness of health assessment and enhance the monitoring and early warning capabilities of abnormal conditions of transmission lines, the method further includes: For each time series feature after feature engineering processing, the pre-built LSTM prediction model is used to predict the features of the corresponding future time period.

[0064] Taking the LSTM prediction model to predict the temperature in the next X hours as an example, the LSTM prediction model structure is designed as shown in Table 2 below, and the loss function is set to use the mean square error (MSE), the optimizer is Adam, and the learning rate and decay rate are preset; other time series features can use the same LSTM prediction model structure features, and for features involving environmental data (such as ice thickness The cross-modal interaction characteristics of the ambient temperature gradient involve the LSTM y prediction model structure layer. An attention mechanism can be added after the LSTM convolution layer 2 to assign a higher weight to environmental data (such as ambient temperature) to improve prediction accuracy.

[0065] Table 2 LSTM prediction model structure

[0066] The specific steps include initializing the input sequence, predicting the temperature in the first hour, updating the input window, predicting the second hour, iterating to N hours, and obtaining the final predicted temperature for the next N hours.

[0067] Starting from the optimization of the random forest regression model, the prediction feature enhancement is used to make the evaluation more forward-looking. Specifically, the features of the future time period predicted for each time series feature after feature engineering processing are fused with the features after feature engineering processing to obtain a new feature vector. The new feature vectors of each collection point currently obtained are input into the retrained random forest regression model respectively, and the health assessment results of each collection point are obtained to replace the original health assessment results of each collection point. Among them, feature enhancement training is required at the same time as feature enhancement. Specifically, the new feature vectors of each collection point marked with health in the historical transmission line are selected as incremental training data to retrain the random forest regression model, so that the retrained random forest regression model can be used for health assessment in the future.

[0068] From the perspective of being able to timely detect potential anomalies in transmission lines and ensure the safe and stable operation of transmission lines, we choose to build an isolation forest anomaly detection model, including initializing isolation trees, recursive segmentation, and forest construction. Taking the data of various dimensional features predicted by the historical LSTM prediction model as input, the isolation forest anomaly detection model is trained to identify normal feature data and abnormal feature data in each dimension, and output the marked abnormal feature data of each dimension. The trained isolation forest anomaly detection model is used to determine whether the various dimensional features predicted by the LSTM prediction model for the corresponding future time period have abnormal feature data tags. Combined with the deep learning algorithm, the preset type of abnormality warning is triggered for the various dimensional feature data with abnormal data tags, assisting in the subsequent maintenance of the transmission line.

[0069] Specifically, a deep learning algorithm can be used to construct a transmission line abnormality warning module. The input of the model is the various dimensional features of the future time period marked with abnormal feature data, and the output triggers a preset type of abnormal warning; for example: for heavy-load line temperature > 80°C or light-load line temperature > 60°C, the isolation forest model is used to identify data that deviates from the normal line range, triggering a conductor temperature warning; for tower inclination angle > 3°, the isolation forest model is used to identify data that deviates from the normal tower inclination range, generating an ice thickness warning.

[0070] In addition, considering the allocation of computing resources, in order to complete the health warning and transmission line abnormality warning more timely, the edge-cloud collaborative computing architecture is deployed to complete the health warning and abnormality warning of the transmission line; including: using the edge computing center to complete the feature engineering processing of multi-dimensional perception data, feature prediction of future time periods, and abnormality warning using the isolation forest detection model, that is, uploading the collected data to the edge computing center around the collection point to complete the corresponding processing and warning; using the cloud computing center to complete the health assessment results using the random forest regression model, that is, performing feature engineering processing on the edge computing center and uploading the predicted features to the cloud computing center to complete the health assessment.

[0071] This embodiment works by introducing an LSTM prediction model to predict features for future time periods, enabling early detection of transmission line trends. The predicted features are then integrated with the original features to re-evaluate health, resulting in more accurate results. Furthermore, an isolation forest anomaly detection model is used to identify abnormal data and trigger early warnings, enabling more timely detection of potential risks to transmission lines and ensuring the safe and stable operation of the power system.

[0072] In a specific embodiment, in order to better adapt to changes in transmission lines in different scenarios, improve the accuracy and pertinence of health assessment, and thus more effectively ensure the safe operation of transmission lines in various complex scenarios, the method further includes: For each collection point on the transmission line, the terrain data of the corresponding location is collected, such as the geographical location and equipment data, such as the tower data and the conductor data.

[0073] Based on the environmental data, terrain data and equipment data of each collection point in the transmission line, the scenario type corresponding to the current collection point is determined, including: normal state scenarios, abnormal state scenarios, including: extreme weather scenarios, complex terrain scenarios, external force damage scenarios, equipment aging scenarios, etc.

[0074] When monitoring the health of transmission lines in normal scenarios, the health assessment method described in the above embodiment can be directly applied. When monitoring the health of transmission lines in abnormal scenarios, further optimization is performed in terms of feature engineering, model structure, and hyperparameter optimization, including: In order to determine that the current acquisition point is an abnormal scene, feature engineering optimization processing is performed on the multi-dimensional perception data of the current acquisition point after preprocessing to replace feature engineering processing, and the feature vectors of each acquisition point are obtained by fusion; the feature engineering optimization processing includes: using a single-modal feature extraction method to obtain image features, various sensor time series features and lidar point cloud features; using a cross-modal interactive feature construction method to obtain cross-modal interactive features constructed based on physical rules that conform to the current abnormal scene, cross-modal interactive features constructed based on correlation analysis rules, and cross-modal interactive features constructed for the current scene, that is, special construction features; for example: in extreme weather scenes, the construction adds atmospheric The intensity-response characteristics of the image factor and the equipment status are constructed, such as: the cross-modal interaction characteristics that characterize the dynamic relationship between ice load and tower stability = (snow thickness × wind speed) / tower inclination; in complex terrain scenarios, the terrain and equipment coupling characteristics are constructed, such as: the cross-modal interaction that quantifies the amplifying effect of terrain on tower stability = tower inclination / terrain slope; in external force damage scenarios, spatiotemporal mutation characteristics are constructed, such as: the cross-modal interaction characteristics that characterize the approach speed of construction machinery = the moving speed of the object detected by the lidar; in equipment aging scenarios, interaction terms that add equipment operating time and key parameters are constructed, such as: the cross-modal interaction characteristics that quantify the aging rate = insulator operating life × leakage current growth rate.

[0075] To determine if the current collection point is an abnormal scenario, a dedicated model can be trained for each abnormal scenario to achieve model and hyperparameter optimization. Specifically, multiple random forest regression models are constructed, each trained using the feature vectors of historical collection points under a specific scenario. Based on the currently determined abnormal scenario, the corresponding random forest regression model is obtained, and the feature vectors of each collection point currently obtained are input into the trained random forest regression model to obtain the corresponding health assessment results for each collection point.

[0076] In addition, each random forest regression model completes the training process using the feature vectors of each historical collection point under a scenario. Different abnormal scenario types can further improve the model structure. For example, for extreme weather scenarios (blizzards, high temperatures, thunderstorms, etc.), weighted sample random forest is used for structural improvement, assigning higher weights to samples during extreme weather periods to strengthen the model's learning of the characteristics of high-risk periods. For complex terrain scenarios (mountains, jungles, canyons, etc.), layered random forest is used for structural improvement. Sub-forests are divided according to terrain type. Each sub-forest only learns data corresponding to the terrain, and the results are finally fused through weighted calculation (sub-forests with more complex terrain have higher weights). For external force damage scenarios (construction machinery approaching, theft, collisions, etc.), incremental random forest is used for structural improvement. The incremental learning structure is fixed with a fixed initial tree structure. Only a small number of trees are added for new damage samples to capture new features. For equipment aging scenarios (insulator aging, conductor fatigue, tower corrosion, etc.), time window random forest is used for structural improvement. The data is divided into time windows, and sub-forests are trained in each window to capture characteristics of different aging stages. The sub-forest of the corresponding window is called during prediction.

[0077] The implementation principle of this embodiment is as follows: considering the characteristics of different transmission line scenarios, the scenario type is determined by collecting terrain data and equipment data; feature engineering optimization processing is performed on abnormal scenarios and a specialized random forest regression model is used to better adapt to changes in transmission lines under different scenarios, thereby improving the accuracy and pertinence of health assessments, thereby more effectively ensuring the safe operation of transmission lines in various abnormal scenarios.

[0078] like Figure 2 As shown, the embodiment of the present application discloses a multi-dimensional perception panoramic monitoring system for transmission lines, specifically including: The transmission line data acquisition module 101 is used to obtain multi-dimensional sensing data of each collection point on the transmission line, including image data of the collection point, time series data of various sensors, lidar point cloud data of the distance to obstacles around the collection point, and environmental data, and perform pre-processing on each collection point; The transmission line data processing module 102 is used to perform feature engineering processing on the pre-processed multi-dimensional sensing data of each collection point, and fuse them to obtain the feature vector of each collection point. The feature engineering processing includes: using a single-modal feature extraction method to obtain image features, various sensor time series features, and lidar point cloud features; using a cross-modal interactive feature construction method to obtain cross-modal interactive features based on physical rules and cross-modal interactive features based on correlation analysis rules; and using a Gini importance screening method to complete feature screening. The transmission line health assessment module 103 is used to construct a random forest regression model; complete the definition of the health of the transmission line and generate a transmission line health label; select the feature vectors of each collection point marked with health in the historical transmission line as training data, input them into the constructed random forest regression model, and obtain a trained random forest regression model; input the feature vectors of each collection point currently obtained into the trained random forest regression model respectively, and obtain the health assessment results of each collection point; compare the health assessment results of each collection node with the preset health threshold, and generate a health warning for the collection node whose health is less than the preset health threshold.

[0079] In a specific embodiment, the system further includes: The transmission line data prediction module 104 is used to predict the features of the corresponding future time period using a pre-built LSTM prediction model for each time series feature after feature engineering processing; The transmission line health assessment optimization module 105 is configured to combine the features of the future time period predicted for each time series feature processed by feature engineering with the features processed by feature engineering to obtain a new feature vector, input the new feature vector of each collection point currently obtained into the retrained random forest regression model, and obtain the health assessment results of each collection point to replace the original health assessment results of each collection point; wherein the new feature vectors of each collection point with health status marked in the historical transmission line are selected as incremental training data to retrain the random forest regression model; The transmission line abnormality warning module 106 constructs an isolation forest anomaly detection model; uses the data of each dimensional feature predicted by the historical LSTM prediction model as input, trains the isolation forest anomaly detection model to identify normal feature data and abnormal feature data of each dimension, and outputs the marked abnormal feature data of each dimension; uses the trained isolation forest anomaly detection model to determine whether the dimensional features predicted by the LSTM prediction model for the corresponding future time period have abnormal feature data tags, and combines with the deep learning algorithm to trigger a preset type of abnormality warning for the dimensional feature data with abnormal data tags.

[0080] In a specific embodiment, the transmission line data acquisition module 101 of the system is further configured to acquire terrain data and equipment data for each acquisition point of the transmission line; determine the scene type corresponding to the current acquisition point based on the environmental data, terrain data, and equipment data of each acquisition point in the transmission line, including: a normal state scene and an abnormal state scene; The transmission line data adaptive processing module 107, upon determining that the current collection point is an abnormal scenario, performs feature engineering optimization processing on the pre-processed multi-dimensional perception data of the current collection point to replace feature engineering processing, and fuses and obtains feature vectors of each collection point; the feature engineering optimization processing includes: using a single-modal feature extraction method to obtain image features, various sensor time series features, and lidar point cloud features; using a cross-modal interactive feature construction method to obtain cross-modal interactive features constructed based on physical rules that conform to the current abnormal scenario, and cross-modal interactive features constructed based on correlation analysis rules; and using a Gini importance screening method to complete feature screening; The transmission line adaptive health assessment module 108 constructs multiple random forest regression models. Each random forest regression model is trained using the feature vectors of each historical collection point under a scenario. The corresponding random forest regression model is obtained according to the currently determined abnormal scenario, and the feature vectors of each collection point currently obtained are input into the trained random forest regression model to obtain the corresponding health assessment results of each collection point.

[0081] The embodiment of the present application also discloses a computer-readable storage medium.

[0082] Specifically, the computer-readable storage medium stores a computer program that can be loaded by a processor and executed such as the above-mentioned multi-dimensional perception panoramic monitoring method for transmission lines. The computer-readable storage medium includes, for example: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0083] The embodiment of the present application also discloses a computer device.

[0084] Specifically, the computer device includes a memory and a processor, and the memory stores a computer program that can be loaded by the processor and execute the above-mentioned multi-dimensional perception panoramic monitoring method for transmission lines.

[0085] The above are all preferred embodiments of the present application and are not intended to limit the scope of protection of this application. Unless otherwise stated, any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features. In other words, unless otherwise stated, each feature is merely an example of a series of equivalent or similar features.

Claims

1. A multi-dimensional perception panoramic monitoring method for power transmission lines, characterized in that: include: For each collection point on the transmission line, multi-dimensional perception data is obtained, including: image data of the collection point, time series data of various sensors, lidar point cloud data of the distance to obstacles around the collection point, and environmental data, and pre-processed for each collection point; Feature engineering is performed on the multi-dimensional perception data of each collection point after preprocessing, and the feature vectors of each collection point are obtained by fusion. The feature engineering process includes: using a single-modal feature extraction method to obtain image features, various sensor time series features, and lidar point cloud features; using a cross-modal interactive feature construction method to obtain cross-modal interactive features based on physical rules and cross-modal interactive features based on correlation analysis rules; and using a Gini importance screening method to complete feature screening. Construct a random forest regression model; complete the definition of transmission line health and generate a transmission line health label; select the feature vectors of each collection point marked with health in the historical transmission line as training data, input it into the constructed random forest regression model, and obtain a trained random forest regression model; The feature vectors of each collection point currently obtained are input into the trained random forest regression model to obtain the health assessment results of each collection point; the health assessment results of each collection node are compared with the preset health threshold, and a health warning is generated for the collection node whose health is less than the preset health threshold.

2. The multi-dimensional perception panoramic monitoring method for power transmission lines according to claim 1 is characterized in that: Also includes: For each time series feature after feature engineering, use the pre-built LSTM prediction model to predict the features of the corresponding future time period; The features of the future time period predicted for each time series feature processed by feature engineering are combined with the features processed by feature engineering to obtain a new feature vector. The new feature vectors of each collection point currently obtained are respectively input into the retrained random forest regression model, and the health assessment results of each collection point are obtained to replace the original health assessment results of each collection point. Among them, the new feature vectors of each collection point marked with health in the historical transmission line are selected as incremental training data to retrain the random forest regression model; Construct an isolation forest anomaly detection model; use the historical data of each dimensional feature predicted by the LSTM prediction model as input, train the isolation forest anomaly detection model to identify the normal feature data and abnormal feature data of each dimension, and output the marked abnormal feature data of each dimension; use the trained isolation forest anomaly detection model to determine whether the dimensional features of the LSTM prediction model corresponding to the future time period have abnormal feature data tags, and combine with the deep learning algorithm to trigger the preset type of abnormal warning for the dimensional feature data with abnormal data tags.

3. The multi-dimensional perception panoramic monitoring method for power transmission lines according to claim 1, characterized in that: Also includes: Collect and obtain terrain data and equipment data at each collection point on the transmission line; Based on the environmental data, terrain data, and equipment data of each collection point in the transmission line, determine the scenario type corresponding to the current collection point, including: normal state scenario and abnormal state scenario; the abnormal state scenario includes: extreme weather scenario, complex terrain scenario, external force damage scenario, and equipment aging scenario; In order to determine that the current acquisition point is an abnormal scene, feature engineering optimization processing is performed on the pre-processed multi-dimensional perception data of the current acquisition point to replace feature engineering processing, and the feature vectors of each acquisition point are obtained by fusion; the feature engineering optimization processing includes: using a single-modal feature extraction method to obtain image features, various sensor time series features, and lidar point cloud features; using a cross-modal interactive feature construction method to obtain cross-modal interactive features constructed based on physical rules that conform to the current abnormal scene, and cross-modal interactive features constructed based on correlation analysis rules; and using a Gini importance screening method to complete feature screening; Construct multiple random forest regression models, each of which is trained using the feature vectors of each historical collection point under a scenario; obtain the corresponding random forest regression model based on the currently determined abnormal scenario, and input the currently obtained feature vectors of each collection point into the trained random forest regression model to obtain the corresponding health assessment results of each collection point.

4. The multi-dimensional perception panoramic monitoring method for power transmission lines according to claim 1, characterized in that: Also includes: Define the transmission line health evaluation based on multi-dimensional indicators, including structural health, electrical health, and environmental health; For each collection point of the transmission line, the status data of the corresponding multi-dimensional perception data collection device is obtained, and the weights of indicators of different dimensions are dynamically adjusted according to the status data of the multi-dimensional perception data collection device; including: separately counting the number of devices with fault risks according to the status data of the corresponding collection devices of the corresponding dimensional perception data of structural health, electrical health and environmental health, and dynamically adjusting the weights of indicators of different dimensions according to the proportion of the number of devices with fault risks counted in different dimensions.

5. The multi-dimensional perception panoramic monitoring method for power transmission lines according to claim 1, characterized in that: The random forest regression model construction includes: Initialize random forest parameters, including the number of decision trees, maximum depth, and minimum number of samples for splitting nodes; Bootstrap sampling is used to extract sample subsets with replacement from the feature vector of the original data, and multiple decision trees are constructed. When each decision tree splits, some features are randomly selected to obtain the final decision trees as training data. The random forest regression model is fitted using the training data. The input of the model is the feature vector obtained from the collection point. A single decision tree is used to predict the output. The mean of the prediction output of all single decision trees is calculated or the K-means clustering method is used to obtain the cluster mean to obtain the overall prediction output of the forest and obtain the health assessment result.

6. The multi-dimensional perception panoramic monitoring method for power transmission lines according to claim 2, characterized in that: Also includes: The edge-cloud collaborative computing architecture is deployed to provide health and anomaly warnings for power transmission lines. This includes: using an edge computing center to perform feature engineering processing of multi-dimensional sensor data, feature prediction for future time periods, and anomaly warning using the isolation forest detection model. The cloud computing center is used to obtain health assessment results using the random forest regression model.

7. The multi-dimensional perception panoramic monitoring method for power transmission lines according to claim 1, characterized in that: Also includes: After completing feature screening using the Gini importance screening method, VIF collinearity test is performed on the screened features to obtain the variance inflation coefficient of each screened feature, and features whose variance inflation coefficient is not greater than the preset variance inflation coefficient are retained.

8. A multi-dimensional perception panoramic monitoring system for power transmission lines, characterized by: include: The transmission line data acquisition module is used to obtain multi-dimensional perception data for each collection point on the transmission line, including image data of the collection point, time series data of various sensors, lidar point cloud data of the distance to obstacles around the collection point, and environmental data, and perform pre-processing on each collection point; The transmission line data processing module is used to perform feature engineering processing on the pre-processed multi-dimensional sensing data of each collection point, and fuse them to obtain the feature vector of each collection point. The feature engineering processing includes: using a single-modal feature extraction method to obtain image features, various sensor time series features, and lidar point cloud features; using a cross-modal interactive feature construction method to obtain cross-modal interactive features based on physical rules and cross-modal interactive features based on correlation analysis rules; and using a Gini importance screening method to complete feature screening. The transmission line health assessment module is used to construct a random forest regression model; complete the definition of transmission line health and generate a transmission line health label; select the feature vectors of each collection point marked with health in the historical transmission line as training data, input the constructed random forest regression model, and obtain a trained random forest regression model; input the currently obtained feature vectors of each collection point into the trained random forest regression model respectively, and obtain the corresponding health assessment results of each collection point; compare the health assessment results of each collection node with the preset health threshold respectively, and generate a health warning for the collection node whose health is less than the preset health threshold.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 7.

10. A computer device, characterized in that: The computer device includes a memory, a processor, and a program stored and executable on the memory, and when the program is executed by the processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images