Power transmission line monitoring and early warning method, system and equipment based on Beidou communication

Abstract

The invention discloses a power transmission line monitoring and early warning method, system and device based on Beidou communication, and relates to the technical field of electric power engineering. The method comprises the following steps: dividing span units, and screening a time sequence dynamic factor and a multi-modal image factor to form a fusion factor set; constructing a fusion Transform time sequence prediction sub-model and a multi-modal feature fusion sub-model, and outputting a fault tendency index; calculating a fault probability index, and dividing risk grade regions; carrying out weight distribution on the fault consequence evaluation factor, and calculating a fault influence index; a dynamic weight distributor is constructed, weight distribution is adjusted based on seasons and real-time meteorological conditions, and an objective function is optimized; and calculating a comprehensive risk value R, and dividing four early warning levels. Through multi-modal time sequence fusion, dynamic weight distribution and a Beidou closed-loop cooperation mechanism, the precision and response efficiency of power transmission line fault early warning are improved, and a solid technical guarantee is provided for safe and stable operation of a power system.

Description

Technical Field

[0001] This invention belongs to the field of power engineering technology, specifically relating to a method, system and equipment for monitoring and early warning of power transmission lines based on Beidou communication. Background Technology

[0002] As the lifeline of the power system, the safe and stable operation of transmission lines directly affects the reliability of power supply. Traditional transmission line monitoring mainly relies on manual inspections, periodic maintenance, and data from single sensors, which has the following core drawbacks:

[0003] Weak time-series prediction capability: Traditional methods often use threshold alarms or simple statistical models, which cannot make long-term time-series predictions of fault tendencies, resulting in delayed early warnings or high false alarm rates.

[0004] Static risk assessment: Fault risk level classification is mostly based on fixed thresholds, without dynamically adjusting weights in conjunction with real-time meteorological conditions and seasonal characteristics, making it difficult to reflect the spatiotemporal heterogeneity of fault probability.

[0005] Low communication and collaboration efficiency: Monitoring of power transmission lines in remote areas relies on public network communication, which has signal blind spots; after the early warning information is issued, the operation and maintenance response lacks a closed-loop management mechanism, and the progress of handling cannot be tracked in real time, resulting in low efficiency in fault handling.

[0006] With the mature application of the BeiDou Navigation Satellite System, its centimeter-level high-precision positioning, short message communication, and differential correction technology have provided new technical support for power transmission line monitoring. However, existing BeiDou-based power transmission line monitoring solutions mostly focus on positioning calibration or single fault detection, without systematically integrating multimodal data fusion, time-series prediction models, and dynamic weight allocation mechanisms, making it difficult to achieve intelligent upgrades from fault detection to risk warning.

[0007] To address the aforementioned problems, this invention proposes a method, system, and device for monitoring and early warning of power transmission lines based on BeiDou communication. Summary of the Invention

[0008] In order to overcome the shortcomings and deficiencies of the existing technologies, the first objective of this invention is to provide a network security situation awareness system based on big data; the second objective of this invention is to provide a power transmission line monitoring and early warning method based on BeiDou communication; and the third objective of this invention is to provide a power transmission line monitoring and early warning device based on BeiDou communication.

[0009] The first objective of this invention is achieved through the following technical solution:

[0010] The method for monitoring and early warning of power transmission lines based on BeiDou communication includes the following steps:

[0011] S1. Identify the fault area and normal area of ​​the transmission line, divide the span unit of the fault area, filter the fault tendency assessment factor in the span unit, and select the training set and test set from the fault area and normal area based on the fault tendency assessment factor.

[0012] S2. Establish a fault analysis model that integrates Transformer time-series prediction, optimize the hyperparameters of the fault analysis model to obtain an optimized analysis model, use the optimized analysis model to output the fault tendency time-series prediction results and classification results, determine the fault tendency and time-series trend analysis results of the gear unit, and label them as the fault tendency index Y.

[0013] S3. Calculate the failure probability index H of the gear unit, and use the failure probability index H to divide the risk level zone of the failure area.

[0014] S4. Filter the fault consequence assessment factors in the gear unit, generate the indicator hierarchy of the fault consequence assessment factors, construct the judgment matrix of the indicator hierarchy, and use the judgment matrix to perform hierarchical analysis on the fault consequence assessment factors to obtain the hierarchical analysis results.

[0015] S5. Calculate the fault impact index V of the gear unit using the hierarchical analysis results;

[0016] S6. Based on the dynamic weighted fusion of the fault probability index H, fault impact index V, and fault tendency index Y, the power transmission line fault monitoring and early warning results are generated, and the time-series early warning information is sent out through the Beidou communication module.

[0017] Preferably, the gear unit for dividing the fault area includes:

[0018] Obtain GIS spatial data of power transmission lines and tower coordinate information;

[0019] The coordinates of the tower are calibrated in real time using the BeiDou positioning system to correct spatial position deviations.

[0020] Based on the calibrated tower coordinates, extract the line segment boundaries between adjacent towers;

[0021] The line segment boundaries were verified for compliance in accordance with the power transmission line design specifications.

[0022] Generate span unit attribute information including span length, tower type, and conductor parameters;

[0023] By integrating attribute information and spatial boundaries, a span unit model is obtained;

[0024] Search for the gear unit corresponding to the gear unit model in the fault area.

[0025] Preferably, the fault tendency assessment factors in the filter unit include:

[0026] Divide the grid cells within the spacing unit;

[0027] Select temporal dynamic factors, meteorological factors, line parameter factors, environmental factors, and multimodal image factors from the raster cells;

[0028] Meteorological factors, line parameter factors, and environmental factors are quantitatively classified to obtain quantitative classification factors.

[0029] Temporal dynamic factors are preprocessed using time series standardization, and deep features are extracted from multimodal image factors using a convolutional neural network.

[0030] Calculate the Pearson coefficient of the quantification grading factor, and remove quantification grading factors whose absolute value of the Pearson coefficient is greater than the preset factor threshold to obtain the remaining grading factors.

[0031] By fusing the remaining grading factors, preprocessed temporal dynamic factors, and multimodal image depth features, a fusion factor set for the span unit is generated, namely the fault tendency assessment factor.

[0032] Preferably, a fault analysis model integrating Transformer time-series prediction is established, including:

[0033] S21. Construct a Transformer-based time series prediction sub-model to process dynamic time series factors and predict the changing trend of dynamic time series factors in a future preset time period.

[0034] S22. Construct a multimodal feature fusion sub-model to fuse the future time-series trend features output by the time-series prediction sub-model, the depth features of multimodal image factors, and the quantized grading factor vector, and input them into the classifier to output the fault tendency classification result, which is then mapped to the fault tendency index Y.

[0035] S23. Train and optimize the hyperparameters of the fault analysis model to obtain the optimized analysis model.

[0036] Preferably, the calculation of the failure probability index H of the gear unit includes:

[0037] Using formula Calculate the probability of failure occurrence, where, This represents the probability of failure of the i-th span unit under extreme weather conditions within a future preset time period. This indicates the maximum intensity of extreme weather events historically monitored for this range unit. This represents the predicted intensity of future extreme weather events by Transformer.

[0038] Based on the probability of failure Using the formula Calculate the failure probability index; where, Let represent the failure probability index of the i-th gear unit, with a value range of [0,1]. Indicates the probability of a failure occurring. This indicates the failure tendency index of the gear unit. This represents the maximum failure tendency index for all gear units.

[0039] Preferably, the fault probability index H, fault impact index V, and fault tendency index Y are fused based on dynamic weights, including:

[0040] S51. Construct a dynamic weight allocator to calculate the dynamic weights of the fault probability index H, fault impact index V, and fault tendency index Y based on seasonal type and real-time weather conditions. , , ;

[0041] S52, Using the formula Calculate the comprehensive risk value R; S53. Based on the comprehensive risk value R, classify the early warning level and generate the transmission line fault monitoring and early warning results;

[0042] After issuing time-series early warning information via the BeiDou communication module, it also includes:

[0043] Build an operation and maintenance collaborative management platform; receive feedback from operation and maintenance personnel on response status, on-site verification, handling progress and processing results within a specified time limit based on early warning information; associate the processing results with the corresponding early warning records to form a closed-loop archive.

[0044] Preferably, the method further includes: periodically reviewing and analyzing the closed-loop archives to calculate the early warning consistency; if the early warning deviation is greater than a set threshold, triggering the model parameter tuning process or the factor screening optimization process, and updating the historical fault dataset on which the dynamic weight allocation depends.

[0045] The second objective of this invention is achieved through the following technical solution:

[0046] The transmission line monitoring and early warning system based on BeiDou communication is used to realize the transmission line monitoring and early warning method based on BeiDou communication. The system includes a spatial data acquisition and span modeling module, a multimodal factor screening and feature fusion module, an intelligent prediction and dynamic evaluation module, and a closed-loop collaboration and optimization module.

[0047] The spatial data acquisition and span modeling module includes a BeiDou high-precision positioning submodule and a span unit modeling submodule. The BeiDou high-precision positioning submodule uses differential correction technology to calibrate tower coordinates, acquires data on transmission line alignment, tower distribution, and topography, extracts the boundaries of adjacent tower segments using ArcGIS, and generates attribute information. The span unit modeling submodule integrates spatial and attribute information to generate a GeoJSON format span unit model, and uses spatial overlay analysis to filter key monitoring span units that intersect fault areas.

[0048] The multimodal factor screening and feature fusion module includes a rasterization factor processing submodule and a multimodal feature extraction submodule. The rasterization factor processing submodule divides the span unit into a raster, screens time-series dynamic factors, meteorological factors, line parameter factors, and environmental factors, quantizes and classifies them using the natural discontinuity method, and removes redundant factors using the Pearson coefficient. The multimodal feature extraction submodule extracts deep features and concatenates them with the time-series dynamic factor time series matrix and the quantized and classified factor vector to form a fusion feature tensor, which is used as the model input.

[0049] The intelligent prediction and dynamic assessment module includes a Transformer time-series prediction submodule, a multimodal fusion classification submodule, and a dynamic weight allocation submodule. The Transformer time-series prediction submodule extracts temporal periodic features through a causal convolutional network, encodes location information using a triangular temporal localization network, and predicts trends using an Encoder-Decoder architecture, outputting a predicted sequence. The multimodal fusion classification submodule concatenates time-series prediction features, image CNN features, and quantized grading factors, inputs them into a random forest classifier, outputs a fault tendency classification, and maps it to a fault tendency index. The dynamic weight allocation submodule optimizes the weight allocation of the fault probability index, fault impact index, and fault tendency index based on seasonal and real-time weather conditions using gradient descent, and combines this with the natural discontinuity method to divide the warning levels into four levels, clearly defining the response timeframe.

[0050] The closed-loop collaboration and optimization module includes a warning information distribution sub-module and a closed-loop management platform sub-module. The warning information distribution sub-module distributes time-series warning information, including the warning level, grade unit number, risk trigger, dynamic weight basis, and handling measures, and synchronizes it to the cloud platform to generate a time-series ledger. The closed-loop management platform sub-module provides warning response status feedback, updates on on-site verification progress, correlation of handling results, and generation of closed-loop archives. It triggers model tuning or factor screening optimization through warning consistency review.

[0051] The third objective of this invention is achieved through the following technical solution:

[0052] A power transmission line monitoring and early warning device based on BeiDou communication is used to implement a power transmission line monitoring and early warning method based on BeiDou communication. The device includes:

[0053] Beidou high-precision positioning components: Beidou receivers integrating differential correction technology, tower coordinate calibration, and extraction of adjacent tower line segment boundaries via ArcGIS;

[0054] Span unit modeling component: Based on compliance verification, it integrates spatial-attribute information to generate a GeoJSON format span unit model, filters key monitoring span units with overlapping fault areas through spatial overlay analysis, and generates attribute information by associating with power equipment management system data;

[0055] Rasterization factor processor: Divides the span unit into a grid, filters time-series dynamic factors, meteorological factors, line parameter factors and environmental factors, uses the natural discontinuity method for quantification and classification, and eliminates redundant factors through Pearson coefficient.

[0056] Multimodal feature extractor: Extracts depth features from UAV imagery and meteorological satellite cloud images, and concatenates them with the time series matrix of temporal dynamic factors and the quantized hierarchical factor vector to form a fusion feature tensor, which is used as the model input;

[0057] Transformer temporal prediction component: Extracts temporal periodic features, combines them with triangular temporal localization network to encode location information, predicts trends, and outputs predicted sequences;

[0058] Multimodal fusion classifier: It concatenates time-series prediction features, image CNN features and quantization grading factors, inputs them into a random forest classifier, outputs fault tendency classification, and maps it to a fault tendency index.

[0059] Dynamic weight allocator: Based on seasonal and real-time weather conditions, the weight allocation of the fault probability index, fault impact index, and fault tendency index is optimized by gradient descent method, and the four-level warning level is divided by natural discontinuity method, and the handling time limit is clearly defined.

[0060] Early warning information distribution component: Distributes time-series early warning information, including early warning level, grade unit number, risk cause, dynamic weight basis and handling measures, and synchronizes it to the cloud platform to generate a time-series ledger;

[0061] Closed-loop management platform components: An operation and maintenance platform that integrates BeiDou communication interconnection to realize early warning response status feedback, on-site verification progress updates, handling result correlation and closed-loop file generation, and trigger model tuning or factor screening optimization through early warning consistency review.

[0062] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0063] 1. This invention calibrates tower coordinates using BeiDou positioning differential technology, combines GIS spatial data to divide span units, and constructs a rasterized model to achieve precise spatial positioning and multi-dimensional monitoring of fault-inducing sources. This method effectively integrates temporal dynamic factors and multimodal unstructured data, eliminates redundant factors through Pearson coefficients and quantifies and classifies them, thereby improving the independence and monitoring accuracy of fault tendency assessment factors.

[0064] 2. This invention achieves a breakthrough in intelligent dynamic risk assessment by constructing a fusion of a Transformer temporal prediction sub-model and a multimodal feature fusion sub-model. The Transformer sub-model extracts the periodic features of temporal dynamic factors through a causal convolutional network and encodes location information using a triangular temporal localization network to accurately predict future temporal trends. The multimodal sub-model extracts image depth features through a CNN, concatenates them with quantized grading factors to form a fusion feature tensor, and outputs a fault tendency index through a random forest classifier to improve prediction accuracy.

[0065] 3. This invention introduces a dynamic weight allocation mechanism based on seasonal and real-time meteorological conditions. It optimizes the weight allocation of the fault probability index, fault impact index, and fault tendency index using gradient descent, making the overall risk value more closely reflect actual operating conditions. Through closed-loop verification and iterative optimization mechanisms, it continuously triggers model tuning or factor selection optimization, improving subsequent early warning accuracy and response efficiency. Attached Figure Description

[0066] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0067] Figure 1 A flowchart of the power transmission line monitoring and early warning method based on BeiDou communication of the present invention is shown;

[0068] Figure 2 The diagram shows a module diagram of the power transmission line monitoring and early warning system based on BeiDou communication according to the present invention. Detailed Implementation

[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more exemplary embodiments. Numerous specific details are provided in the following description to give a full understanding of exemplary embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced with one or more of the specific details omitted, or other methods, components, steps, etc., can be employed. In other instances, well-known structures, methods, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0071] Example 1:

[0072] See Figure 1 As shown in the figure, the transmission line monitoring and early warning method based on Beidou communication in this embodiment has the following process:

[0073] S1. Identify the fault area and normal area of ​​the transmission line, divide the span unit of the fault area, filter the fault tendency assessment factors (including time series factors and multimodal factors) in the span unit, and select the training set and test set from the fault area and normal area based on the fault tendency assessment factors.

[0074] In this embodiment of the invention, the fault area refers to the area where transmission line faults have occurred in the past, and the normal area refers to the area where transmission line faults have not occurred in the past and the operating status is stable. Transmission line faults include, but are not limited to, ice galloping, conductor breakage, tower tilting, and insulator flashover.

[0075] Furthermore, this embodiment of the invention uses span units to divide the fault area, thereby monitoring and analyzing the inducing sources of transmission line faults. A span unit refers to a transmission line segment unit defined by two adjacent towers. It is the basic unit for stress analysis, environmental impact response, and fault occurrence of the transmission line, and is directly related to the mechanical characteristics of the transmission line, surrounding environmental conditions, and the layout of monitoring points.

[0076] The method for dividing the fault area into gear units includes:

[0077] The process involves acquiring GIS spatial data of transmission lines and tower coordinates. The GIS spatial data includes, but is not limited to, transmission line routing, tower distribution, and topographic data. The positioning accuracy of the BeiDou positioning system is optimized to the centimeter level through differential correction technology to ensure the accuracy of span unit boundary division. Optionally, the process of acquiring GIS spatial data of transmission lines and tower coordinates can be achieved through the interface call of the power industry GIS platform.

[0078] The coordinates of the tower are calibrated in real time by using the BeiDou positioning system to correct spatial position deviations. The positioning point data at the top of the tower is collected by the BeiDou receiver and combined with the differential signal of the reference station to eliminate errors such as satellite clock error and ionospheric delay.

[0079] Based on the calibrated tower coordinates, the line segment boundaries between adjacent towers are extracted using ArcGIS software;

[0080] The compliance verification of the line segment boundaries is carried out in accordance with the design specifications for power transmission lines. The compliance verification process refers to comparing the extracted span length with the design span range and eliminating abnormal boundary data caused by terrain obstruction.

[0081] Generate span unit attribute information that includes span length, tower type, and conductor parameters; the process of generating span unit attribute information refers to associating data such as tower type, conductor cross-section, and installation year in the power equipment management system;

[0082] By fusing attribute information and spatial boundaries, a span unit model is obtained; the span units corresponding to the span unit model are queried in the fault area; the process of fusing attribute information and spatial boundaries refers to encapsulating data in GeoJSON format to form a span unit model containing spatial location and attribute information; the process of querying the corresponding span units in the fault area refers to selecting span units that intersect with the fault area as key monitoring objects through spatial overlay analysis; this embodiment of the invention filters fault tendency assessment factors in span units to screen key influencing factors that can be used for early warning of transmission line fault tendency, where time series factors are dynamic monitoring data with time series characteristics, and multimodal factors include unstructured data such as UAV imagery and meteorological satellite cloud images.

[0083] Fault tendency assessment factors in the filter unit include:

[0084] Divide the span unit into grid units; select the time-series dynamic factors (conductor sag change time series, tower tilt angle time series, conductor vibration frequency time series, Beidou positioning drift time series), meteorological factors, line parameter factors, environmental factors and multimodal image factors (UAV inspection icing image, meteorological satellite cloud image, vegetation encroachment image) in the grid units.

[0085] Meteorological factors, line parameter factors, and environmental factors are quantified and classified to obtain quantified classification factors; time-series dynamic factors are subjected to time-series standardization preprocessing, including sampling frequency unification and missing value interpolation; deep features of multimodal image factors are extracted using a convolutional neural network (CNN); the Pearson coefficient of the quantified classification factors is calculated using the following formula:

[0086] ;in, The Pearson coefficient is represented by cov(x,y), and cov(x,y) represents the covariance between the quantization factor of x and the quantization factor of y. The standard deviation of the quantification factor for x represents the quantification factor. This represents the standard deviation of the y-quantification grading factor. This represents the mean of the quantification factor for x. This represents the mean of the quantization factor of y. express The mathematical expectation of x, where x represents the quantization grading factor and y represents the quantization grading factor other than x;

[0087] Quantitative grading factors with absolute values ​​of Pearson coefficients greater than preset factor thresholds are removed to obtain the remaining grading factors; the median of the remaining grading factors is calculated; the remaining grading factors, the preprocessing results of time-series dynamic factors, and the depth features of multimodal images are fused to generate a fusion factor set for span units, namely the fault tendency assessment factors; among them, meteorological factors include wind speed time series, icing thickness time series, ambient temperature time series, relative humidity time series, and precipitation intensity time series; line parameter factors include conductor type, span length, tower type, and insulator aging degree; and environmental factors include vegetation coverage, terrain slope, distance from pollution source, and lightning activity frequency.

[0088] Among them, grid cell refers to dividing the span cell space into a regular grid, each grid is called a cell. The preset factor threshold refers to the threshold of the absolute value of the Pearson coefficient, which aims to reduce data redundancy and improve the relative independence between fault tendency evaluation factors. The correlation between evaluation factors is evaluated by calculating the Pearson correlation coefficient, and factors with an absolute value of correlation coefficient greater than 0.5 are eliminated to avoid multicollinearity. The process of CNN extracting multimodal image depth features refers to using the ResNet50 network to perform convolution operations on UAV images and outputting a 2048-dimensional image feature vector to represent visual information such as icing thickness and insulator contamination level. Optionally, the process of dividing the grid cells in the range unit, for example, converting all fault tendency evaluation factors into factors within a 5m×5m grid cell, and quantifying and classifying each factor to obtain quantified classification factors, refers to using the natural discontinuity method to quantify and classify multiple factors in the grid cell (classifying icing thickness into four levels: no icing, light icing, moderate icing, and heavy icing). The process of fusing the remaining classification factors, temporal dynamic factors, and multimodal image features refers to concatenating the time series matrix (length T×dimension D) of the temporal dynamic factors, the quantized classification factor vector (dimension M), and the image feature vector (dimension N) into a fused feature tensor (T×(D+M+N)), which serves as the input to the subsequent temporal prediction model. Optionally, the process of selecting training and test sets from faulty and normal regions based on fault tendency assessment factors is as follows: To construct a time-series prediction-classification fusion model, it is necessary to select faulty positive samples and normal negative samples. The positive samples are the fusion data of time-series dynamic factors, multimodal image factors, and quantitative grading factors of the faulty region span unit (including time-series data of the T time period before the fault occurred), and the negative samples are the same-dimensional fusion data of the normal region span unit. The sample set is divided into training and test sets in a 7:3 ratio. The training set contains historical time-series sequences and corresponding labels (faulty / normal), and the test set is used to verify the time-series prediction and classification accuracy of the model.

[0089] S2. Establish a fault analysis model that integrates Transformer time-series prediction, optimize the model for hyperparameters to obtain an optimized analysis model, and use the optimized analysis model to output fault tendency time-series prediction results and classification results. Analyze the accuracy of the results, determine the fault tendency and time-series trend analysis results of the gear unit, and label them as the fault tendency index Y.

[0090] S21. Construct a time series prediction sub-model based on Transformer to handle dynamic time series factors.

[0091] In one embodiment of the present invention, a time-series prediction sub-model based on Transformer is constructed, comprising:

[0092] Obtain the time-series dynamic factors (time series of conductor sag changes, time series of tower tilt angles, etc.) from the fusion factor set, denoted as time series data X=[x1,x2,...,x] t ], where t is the time step;

[0093] The temporal dynamic factors are input into the first sub-network (causal convolutional network + triangular temporal localization network) of the temporal prediction sub-model for preprocessing to obtain global temporal features;

[0094] The encoder model obtains multiple local time series features from global time series features according to a preset time period, and determines the probability distribution of the autocorrelation degree of each local time series feature relative to the global time series features;

[0095] The Decoder model can predict the trend of changes in time-series dynamic factors (such as the predicted value of sag change and the predicted value of tower tilt angle) over a preset time period based on the probability distribution of autocorrelation.

[0096] The preprocessing process of the first sub-network includes: inputting the temporal dynamic factors into the causal convolutional network, compressing the temporal data through causal convolution (using only historical time data), extracting periodic features (such as the daily variation period of sag and the growth period of icing thickness), and obtaining compressed data X'=[x1',x2',...,x t The compressed data is input into a triangular temporal localization network, and the temporal location information of the compressed data is determined by the following formula to obtain the global temporal features:

[0097] ;

[0098] ;

[0099] Where pos represents the relative position of the temporal location information to be encoded in the time series, i represents the dimension of the time series vector, and d represents the dimension of the global temporal feature in the Encoder model;

[0100] The encoder model processing includes: dividing the global temporal features into multiple local temporal features according to a preset time period; performing a masking operation on the global temporal features based on the temporal position information of each local temporal feature; and determining the location of the data unit in the global temporal features that is related to each local temporal feature for calculating the autocorrelation degree according to the following formula:

[0101] ;

[0102] ;

[0103] ;

[0104] ;

[0105] Where Q, K, and V represent the query matrix, key matrix, and value matrix, respectively. , , This is the corresponding weight matrix. is the dimension of the key vector.

[0106] Based on location, the autocorrelation degree of corresponding data units in each local temporal feature and the global temporal feature is calculated to generate a self-attention weight distribution. The Decoder model prediction process includes: inputting the self-attention weight distribution into the Decoder model, combining the future temporal location information generated by the triangular temporal localization network, fusing the global features output by the Encoder using a cross-attention mechanism, and predicting the change value of temporal dynamic factors within a preset time period. The predicted sequence of sag change within the preset time period is Ŷ=[ŷ1,ŷ2,...,ŷ h ], h is the prediction time step.

[0107] S22. Constructing a multimodal feature fusion sub-model to fuse temporal prediction results with non-temporal features. In one embodiment of the present invention, constructing a multimodal feature fusion sub-model includes: concatenating the future temporal trend features (depth features of Ŷ) output by the Transformer temporal prediction sub-model, the CNN features of multimodal image factors, and the quantized grading factor vector to obtain fused features. The fused feature F is input into a random forest classifier, which outputs the fault tendency classification result (extremely high fault tendency / high fault tendency / medium fault tendency / low fault tendency), and is mapped to the fault tendency index Y (value range [0,1]). The time series trend prediction result and the classification result are fused to obtain the fault tendency and time series trend analysis results.

[0108] S23. Model Training and Hyperparameter Optimization In one embodiment of the present invention, model training and hyperparameter optimization include: constructing a training loss function and fusing temporal prediction loss and classification loss.

[0109] ;in, , These are the weighting coefficients. For time series prediction loss, For classification loss.

[0110] The time series prediction loss is calculated using the following formula:

[0111] ;

[0112] in, This represents the predicted dynamic factor value at time position i. This represents the actual dynamic factor value at time position i.

[0113] The classification loss uses cross-entropy loss; the initial analysis model is divided into subspaces to be optimized (number of attention heads in Transformer, number of encoder layers, number of decision trees in random forest, etc.); the hyperparameters to be optimized are uniformly sampled, and the posterior distribution of the objective function (L) is analyzed; based on the posterior distribution, the target subspace is identified using the sampling function AC(z)=μ(z)+kσ(z) (μ(z) is the mean of the posterior distribution, σ(z) is the variance, and k is the tuning parameter), and the hyperparameters are optimized using the Bayesian optimization algorithm to obtain the optimized analysis model. S24, Result Accuracy Analysis In one embodiment of the present invention, the accuracy of the analysis results includes: for the time series prediction results, calculating the root mean square error RMSE to evaluate the prediction accuracy; for the classification results, constructing a confusion matrix and calculating the AUC value of the ROC curve to evaluate the classification accuracy; when RMSE≤ preset time series threshold and AUC≥0.85, the model accuracy is determined to meet the requirements, and the fault tendency index Y and future time series trend of each grade unit are output; if the accuracy does not meet the requirements, return to S1 to re-select factors or expand the time series sample set, and retrain the model.

[0114] S3. Calculate the fault probability index (integrated with time-series prediction trends, denoted as fault probability index H) of the span unit, and use the fault probability index to divide the fault area into risk level zones. This embodiment of the invention achieves dynamic fault probability evaluation by calculating the fault probability index and combining it with future time-series trends predicted by Transformer (such as future icing thickness growth trends and wind speed enhancement trends). In one embodiment of the invention, calculating the fault probability index of the span unit includes: calculating the probability of fault occurrence (integrated with time-series prediction trends) using the following formula:

[0115] ;in, This represents the probability of failure of the i-th span unit under extreme weather conditions within a future preset time period. This indicates the maximum intensity of extreme weather events historically monitored for this range unit. This represents the predicted intensity of future extreme weather events by Transformer.

[0116] The failure probability index is calculated using the following formula based on the probability of failure occurrence: ;in, Let represent the failure probability index of the i-th gear unit, with a value range of [0,1]. Indicates the probability of a failure occurring. This indicates the failure tendency index of the gear unit. This represents the maximum failure tendency index for all gear units.

[0117] Optionally, the process of dividing risk level zones using the failure probability index refers to dividing the failure probability index into four levels using the natural discontinuity method, corresponding to extremely high risk zone, high risk zone, medium risk zone, and low risk zone, and generating a risk level zoning map that integrates time-series trends.

[0118] S4. Screen the fault consequence assessment factors in the span unit, generate the hierarchical structure of the fault consequence assessment factors, construct the judgment matrix of the hierarchical structure, and use the judgment matrix to perform hierarchical analysis on the fault consequence assessment factors to obtain the hierarchical analysis results. This embodiment of the invention provides a weighting basis for calculating the fault impact index by screening fault consequence assessment factors and performing hierarchical analysis, while also adjusting the analysis logic based on the fault occurrence time predicted by time series. The specific process includes: screening fault consequence assessment factors in the span unit, including: selecting line importance factors, economic loss factors, repair difficulty factors, and social impact factors; quantifying and classifying each factor to obtain fault consequence assessment factors; wherein, the line importance factor includes whether it is a backbone line, load level, and power supply range; the economic loss factor includes line repair cost, power outage economic loss, and surrounding facility loss; the repair difficulty factor includes terrain complexity, traffic accessibility, and spare parts storage distance; and the social impact factor includes surrounding population density, number of important users, and estimated power outage duration (combined with the fault occurrence time correction based on time series prediction). Generate a hierarchical structure of indicators: Decompose decision elements into a target layer (evaluation results of fault impact), a criterion layer (importance of the line, economic loss, repair difficulty, social impact), and a scheme layer (evaluation factors of each fault consequence); Construct a judgment matrix: Compare the importance of each criterion layer relative to the target layer and each scheme layer relative to the criterion layer pairwise using expert scoring to generate a judgment matrix; Hierarchical analysis: Calculate the eigenvectors and maximum eigenvalues ​​of the judgment matrix using the Analytic Hierarchy Process (AHP), and perform a consistency test (CR < 0.1). After passing the test, obtain the weights of each factor. S5. Calculate the fault impact index (denoted as fault impact index V) of the span unit using the hierarchical analysis results. Determine the fault consequence assessment results of the fault area based on the fault impact index and time series trend. This embodiment of the invention calculates the fault impact index by weighted summation, introduces the protection level reduction coefficient, and combines it with the predicted fault occurrence time for adjustment, making the results more realistic. The specific process includes: obtaining the indicator weight values ​​in the hierarchical analysis results; calculating the fault impact index using the following formula based on the weight values:

[0119] ;

[0120] in, This represents the fault impact index of the i-th gear unit, with a value range of [0,1]. This represents the weight of the j-th failure consequence assessment factor. This represents the standardized value of the j-th factor. This represents the protection level reduction factor. The baseline values ​​for the protection level reduction factor are set as follows: F = 0.2 for a unit with BeiDou high-precision monitoring + active de-icing + intelligent inspection; F = 0.5 for a unit with BeiDou monitoring + passive protection; F = 0.8 for a unit with only basic BeiDou positioning; and F = 1.0 for a unit without protection.

[0121] S51. Construct a dynamic weight allocator to calculate the dynamic weights of H, V, and Y based on seasonal and real-time weather conditions. This embodiment of the invention introduces a dynamic weight allocation mechanism to adjust the weights of each risk factor according to the seasonal type and real-time weather conditions. The specific process includes:

[0122] S511. Determine the weighting factors and correlation discriminant indices. Select seasonal type and real-time weather type as weighting factors, and define the weather-season correlation discriminant index:

[0123] ;

[0124] in, This represents the correlation coefficient of the k-th type of influencing factor (seasonal / meteorological). This represents the weight percentage of the m-th sub-influence factor. The correlation coefficient between the k-th type of influence factor and the m-th sub-influence factor satisfies the following condition: .

[0125] S512. Establish a dynamic weight optimization solution model with the goal of minimizing risk assessment error. Combining relevant discrimination index constraints and weight lower bound constraints, construct the weight optimization model: Objective function:

[0126] ;

[0127] Constraints:

[0128] Relevant discrimination constraints: ; ;

[0129] Weights and constraints: ;

[0130] Weight lower bound constraint: , , .

[0131] in, These are the weighting coefficients for the failure probability index. The weighting coefficients for the fault impact index are... The weighting coefficients for the failure tendency index are... This is a discriminant index for the correlation between seasonal / meteorological factors and failure probability. The correlation index between seasonal / meteorological factors and the impact of faults. This is a discriminant index for the correlation between seasonal / meteorological factors and fault tendency. This is the baseline value for the weighting.

[0132] The above model is solved iteratively using the gradient descent method, minimizing the sum of squared residuals between the actual risk value and the calculated value, thus obtaining the optimal dynamic weights. , , .

[0133] S513. The dynamic weight iteration update mechanism refers to the error trend judgment logic in Implementation Example B, setting the weight update cycle to 1 hour. Each update is performed as follows:

[0134] Calculate the risk assessment error under the current weight. ;

[0135] Define error trend factor ;in, The average error of the first three iterations is T. If T > 0.1, meaning the error is increasing, return to S512 to recalculate the weights. If T ≤ 0.1, meaning the error is decreasing or stable, maintain the current weights until the next cycle.

[0136] S6. Based on the dynamic weighted fusion of H, V, and Y, generate transmission line fault monitoring and early warning results, and send out time-series early warning information through the Beidou communication module. This embodiment of the invention uses a dynamic weighted summation formula to calculate the comprehensive risk value. The specific process includes:

[0137] 1. Calculation of comprehensive risk value: A dynamic weighted summation model is adopted, and the formula is as follows:

[0138] ;

[0139] Where R is the overall risk value, and its value ranges from [0,1]. , , The dynamic weights are obtained for S51, where H, V, and Y are the fault probability index, fault impact index, and fault tendency index, respectively.

[0140] 2. Early Warning Level Classification: The comprehensive risk value R is divided into four early warning levels using the natural discontinuity method, and the response time limit is determined by combining the time series forecast trend.

[0141] Level I (Extremely High Risk, R≥0.8): Failure may occur within the next 6 hours, requiring immediate shutdown and maintenance;

[0142] Level II (High Risk, 0.6≤R<0.8): The probability of failure increases within the next 12 hours, requiring on-site verification within 2 hours;

[0143] Level III (Medium risk, 0.3≤R<0.6): Potential for malfunction exists within the next 24 hours, requiring close monitoring within 24 hours;

[0144] Level IV (Low risk, R < 0.3): Routine monitoring and scheduled inspections.

[0145] 3. Warning Information Issuance and Ledger Generation: Warning information is issued to the operation and maintenance terminal through the Beidou short message communication module. The content includes the warning level, the range unit number, the risk cause, the basis for dynamic weight allocation, the prediction of future fault trends and the suggested handling measures. At the same time, it is synchronized to the cloud management platform to generate a time-series warning ledger, which records the weight iteration process and the risk change curve.

[0146] 4. Construction of a Collaborative Closed-Loop for Early Warning and Response: Establish an operation and maintenance collaborative management platform for BeiDou communication interconnection. After receiving early warning information, operation and maintenance personnel must provide feedback on the initial response status within the time limit specified by the platform (within 30 minutes for Level I early warning, within 1 hour for Level II early warning, within 6 hours for Level III early warning, and within 24 hours for Level IV early warning); after on-site verification, upload the on-site verification situation and the proposed response measures; during the implementation of the response, update the response progress in real time; after the response is completed, submit the processing results, and the platform will automatically associate the corresponding early warning records to form a closed-loop archive.

[0147] 5. Closed-loop verification and iterative optimization: The platform regularly reviews and analyzes the closed-loop files, compares the actual handling results with the Transformer time-series prediction results and the basis for dynamic weight allocation, and calculates the early warning consistency. If the early warning deviation exceeds the set threshold, the platform automatically triggers the model parameter tuning process (returning to S23 to re-optimize hyperparameters) or factor screening optimization (returning to S1 to re-screen and analyze factors), while updating the historical fault dataset of dynamic weight allocation to improve the accuracy of subsequent early warnings.

[0148] The beneficial effects of this embodiment are as follows: by using BeiDou centimeter-level positioning, multimodal time series fusion prediction and dynamic weight allocation technology, it is possible to achieve accurate source location of transmission line faults, intelligent prediction of time series trends and dynamic risk assessment. Combined with a closed-loop collaborative management mechanism, it can improve the accuracy of early warning and response efficiency, and ensure the safe and stable operation of the power system.

[0149] Example 2:

[0150] See Figure 2 As shown, the transmission line monitoring and early warning system based on BeiDou communication in this embodiment includes: a spatial data acquisition and span modeling module, a multimodal factor screening and feature fusion module, an intelligent prediction and dynamic evaluation module, and a closed-loop collaboration and optimization module.

[0151] The spatial data acquisition and gap modeling module includes a BeiDou high-precision positioning submodule and a gap unit modeling submodule.

[0152] The BeiDou high-precision positioning submodule utilizes differential correction technology to achieve centimeter-level tower coordinate calibration. It combines the GIS platform interface to obtain data on transmission line routes, tower distribution, and topography. Through ArcGIS, it extracts the boundaries of adjacent tower segments and generates attribute information including span length, tower type, and conductor parameters.

[0153] Span unit modeling submodule: Based on compliance verification (removing abnormal data due to terrain obstruction), it integrates spatial-attribute information to generate a GeoJSON format span unit model, and filters key monitoring span units with intersection of fault areas through spatial overlay analysis.

[0154] The multimodal factor screening and feature fusion module includes a rasterization factor processing submodule and a multimodal feature extraction submodule.

[0155] Rasterization factor processing submodule: Divide the span unit into 5m×5m grids, filter time-series dynamic factors (sag change, tower tilt, etc.), meteorological factors, line parameter factors and environmental factors, use the natural discontinuity method to quantify and classify (e.g., icing thickness is divided into four levels), and eliminate redundant factors through Pearson coefficient.

[0156] Multimodal feature extraction submodule: Utilizes ResNet50 network to extract 2048-dimensional depth features from UAV images and meteorological satellite cloud images, and concatenates them with the time series matrix of time-series dynamic factors and the quantized hierarchical factor vector to form a fusion feature tensor, which serves as the model input.

[0157] The intelligent prediction and dynamic evaluation module includes a Transformer time-series prediction submodule, a multimodal fusion classification submodule, and a dynamic weight allocation submodule.

[0158] The Transformer temporal prediction submodule extracts temporal periodic features through a causal convolutional network, encodes location information using a triangular temporal localization network, and uses an Encoder-Decoder architecture to predict future trends such as sag changes and icing growth, outputting a predicted sequence.

[0159] Multimodal fusion classification submodule: It concatenates time-series prediction features, image CNN features and quantization grading factors, inputs them into a random forest classifier, outputs fault tendency classification (extremely high / high / medium / low), and maps it to the fault tendency index Y.

[0160] Dynamic weight allocation submodule: Based on seasonal and real-time weather conditions, the weight allocation of the fault probability index H, fault impact index V, and fault tendency index Y is optimized by using the gradient descent method. Combined with the natural discontinuity method, four warning levels are divided, and the handling time limit is specified.

[0161] The closed-loop collaboration and optimization module includes a warning information distribution sub-module and a closed-loop management platform sub-module.

[0162] Early warning information dissemination submodule: Time-series early warning information is disseminated through the BeiDou short message communication module, including early warning level, range unit number, risk cause, dynamic weight basis and handling measures, and is synchronized to the cloud platform to generate a time-series ledger.

[0163] Closed-loop management platform sub-module: Build an operation and maintenance platform for BeiDou communication interconnection to realize early warning response status feedback, on-site verification progress update, handling result correlation and closed-loop file generation, and trigger model tuning or factor screening optimization through early warning consistency review.

[0164] The beneficial effects of this embodiment are as follows: This system achieves accurate source location, intelligent risk assessment and efficient collaborative handling of transmission line faults through BeiDou centimeter-level positioning, multimodal time series fusion prediction, dynamic weight allocation and closed-loop collaborative management, thereby improving early warning accuracy and response efficiency and ensuring the safe and stable operation of the power system.

[0165] Example 3:

[0166] The transmission line monitoring and early warning equipment based on BeiDou communication implemented in this invention includes:

[0167] Beidou high-precision positioning component: Beidou receiver integrating differential correction technology to achieve centimeter-level tower coordinate calibration, obtain transmission line spatial data through GIS platform interface, and extract the boundary of adjacent tower segments through ArcGIS.

[0168] Span unit modeling component: Based on compliance verification, it removes abnormal data caused by terrain obstruction, integrates spatial-attribute information to generate a GeoJSON format span unit model, filters key monitoring span units with overlapping fault areas through spatial overlay analysis, and generates attribute information by associating with power equipment management system data.

[0169] Rasterization factor processor: Divide the span unit into 5m×5m grids, filter time-series dynamic factors (sag change, tower tilt, etc.), meteorological factors, line parameter factors and environmental factors, use the natural discontinuity method for quantification and classification, and eliminate redundant factors through Pearson coefficient, i.e., the absolute value of the correlation coefficient is >0.5.

[0170] Multimodal feature extractor: Deploy ResNet50 network to extract 2048-dimensional depth features from UAV images and meteorological satellite cloud images, and concatenate them with time series matrix of time-series dynamic factors and quantized hierarchical factor vectors to form a fusion feature tensor, which is used as the model input.

[0171] Transformer temporal prediction component: Extracts temporal periodic features through causal convolutional networks, encodes location information by combining triangular temporal localization networks, and uses an Encoder-Decoder architecture to predict future trends such as sag changes and icing growth, outputting a predicted sequence.

[0172] Multimodal fusion classifier: It concatenates time-series prediction features, image CNN features and quantization grading factors, inputs them into a random forest classifier, outputs a fault tendency classification (extremely high / high / medium / low), and maps it to a fault tendency index Y.

[0173] Dynamic weight allocator: Based on seasonal and real-time weather conditions, the weight allocation of the fault probability index H, fault impact index V, and fault tendency index Y is optimized by gradient descent method. Combined with the natural discontinuity method, four warning levels are divided and the handling time limit is specified.

[0174] Early warning information distribution component: Time-series early warning information is distributed through the Beidou short message communication module, including early warning level, range unit number, risk cause, dynamic weight basis and handling measures, and is synchronized to the cloud platform to generate time-series ledger.

[0175] Closed-loop management platform components: An operation and maintenance platform integrating BeiDou communication interconnection to realize early warning response status feedback, on-site verification progress updates, handling result correlation and closed-loop file generation, and trigger model tuning or factor screening optimization through early warning consistency review.

[0176] The beneficial effects of this embodiment are as follows: This device achieves accurate source location, intelligent risk assessment and efficient collaborative handling of transmission line faults through BeiDou centimeter-level positioning, multimodal time series fusion prediction, dynamic weight allocation and closed-loop collaborative management, thereby improving early warning accuracy and response efficiency and ensuring the safe and stable operation of the power system.

[0177] All formulas in this invention are dimensionless and calculated numerically. The preset parameters in the formulas can be set by those skilled in the art according to the actual situation.

[0178] The weighting coefficients of this invention are used to measure the degree of influence of different factors or variables on a certain outcome or decision. The weighting coefficient is defined as the numerical value assigned to each factor when comparing and evaluating multiple factors, reflecting their importance or priority. These weighting coefficients can be determined according to specific circumstances and needs, and are usually jointly formulated and confirmed by professionals or relevant stakeholders. By reasonably setting the weighting coefficients, programs or systems can be helped to make decisions or predictions more accurately.

[0179] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0180] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementations. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for monitoring and early warning of power transmission lines based on BeiDou communication, characterized in that, Includes the following steps: S1. Identify the fault area and normal area of ​​the transmission line, divide the span unit of the fault area, filter the fault tendency evaluation factor in the span unit, and select the training set and test set from the fault area and normal area based on the fault tendency evaluation factor. S2. Establish a fault analysis model that integrates Transformer time-series prediction, optimize the fault analysis model by performing hyperparameter optimization, obtain an optimized analysis model, use the optimized analysis model to output fault tendency time-series prediction results and classification results, determine the fault tendency and time-series trend analysis results of the gear unit, and mark them as fault tendency index Y. S3. Calculate the failure probability index H of the gear unit, and use the failure probability index H to divide the risk level zones in the failure area. S4. Screen the fault consequence assessment factors in the gear unit, generate the indicator hierarchy structure of the fault consequence assessment factors, construct the judgment matrix of the indicator hierarchy structure, and use the judgment matrix to perform hierarchical analysis on the fault consequence assessment factors to obtain the hierarchical analysis results. S5. Calculate the fault impact index V of the gear unit using the hierarchical analysis results; S6. Based on the dynamic weight fusion of the fault probability index H, the fault impact index V, and the fault tendency index Y, generate the transmission line fault monitoring and early warning results, and send out time-series early warning information through the Beidou communication module.

2. The method for monitoring and early warning of power transmission lines based on BeiDou communication according to claim 1, characterized in that, In step S1, the gear spacing unit for dividing the fault area includes: Obtain GIS spatial data of power transmission lines and tower coordinate information; The coordinates of the tower are calibrated in real time using the BeiDou positioning system to correct spatial position deviations. Based on the calibrated tower coordinates, extract the line segment boundaries between adjacent towers; The boundary of the line segment was verified for compliance in accordance with the design specifications for power transmission lines; Generate span unit attribute information including span length, tower type, and conductor parameters; By integrating the attribute information with the spatial boundary, a range unit model is obtained; In the fault region, query the gear unit corresponding to the gear unit model.

3. The method for monitoring and early warning of power transmission lines based on BeiDou communication according to claim 1, characterized in that, In step S1, the fault tendency assessment factor in the screening range unit includes: Divide the grid units in the aforementioned spacing unit; Select the temporal dynamic factors, meteorological factors, line parameter factors, environmental factors, and multimodal image factors from the grid cells; The meteorological factors, line parameter factors, and environmental factors are quantified and classified to obtain quantified classification factors. The time-series dynamic factors are subjected to time-series standardization preprocessing, and the multimodal image factors are subjected to depth features extracted through a convolutional neural network. Calculate the Pearson coefficient of the quantization grading factor, and remove quantization grading factors whose absolute value of the Pearson coefficient is greater than a preset factor threshold to obtain the remaining grading factors. The remaining grading factors, the preprocessed temporal dynamic factors, and the multimodal image depth features are fused to generate the fusion factor set of the range unit, i.e., the fault tendency assessment factor.

4. The method for monitoring and early warning of power transmission lines based on BeiDou communication according to claim 1, characterized in that, In step S2, establishing the fault analysis model that integrates Transformer time-series prediction includes: S21. Construct a Transformer-based time series prediction sub-model to process dynamic time series factors and predict the changing trend of dynamic time series factors in a future preset time period. S22. Construct a multimodal feature fusion sub-model to fuse the future temporal trend features output by the temporal prediction sub-model, the depth features of multimodal image factors, and the quantized grading factor vector, and input them into the classifier to output the fault tendency classification result, which is then mapped to the fault tendency index Y. S23. Train and optimize the hyperparameters of the fault analysis model to obtain the optimized analysis model.

5. The method for monitoring and early warning of power transmission lines based on BeiDou communication according to claim 1, characterized in that, In step S3, calculating the failure probability index H of the gear unit includes: Using formula Calculate the probability of failure occurrence, where, This represents the probability of failure of the i-th span unit under extreme weather conditions within a future preset time period. This indicates the maximum intensity of extreme weather events historically monitored for this range unit. This represents the predicted intensity of future extreme weather events by Transformer. Based on the probability of the failure Using the formula Calculate the failure probability index; where, Let represent the failure probability index of the i-th gear unit, with a value range of [0,1]. Indicates the probability of a failure occurring. This indicates the failure tendency index of the gear unit. This represents the maximum failure tendency index for all gear units.

6. The method for monitoring and early warning of power transmission lines based on BeiDou communication according to claim 1, characterized in that, In step S6, the dynamic weighted fusion of the fault probability index H, the fault impact index V, and the fault tendency index Y includes: S51. Construct a dynamic weight allocator to calculate the dynamic weights of the fault probability index H, the fault impact index V, and the fault tendency index Y based on the seasonal type and real-time weather conditions. , , ; S52, Using the formula Calculate the comprehensive risk value R; S53. Divide the early warning level according to the comprehensive risk value R and generate the transmission line fault monitoring early warning result; After the time-series early warning information is sent through the BeiDou communication module, it also includes: Build an operation and maintenance collaborative management platform; receive response status, on-site verification status, handling progress and processing results from operation and maintenance personnel within a specified time limit based on the early warning information; associate the processing results with the corresponding early warning records to form a closed-loop archive.

7. The method for monitoring and early warning of power transmission lines based on BeiDou communication according to claim 6, characterized in that, The method further includes: periodically reviewing and analyzing the closed-loop archives to calculate the early warning consistency; if the early warning deviation is greater than a set threshold, triggering the model parameter tuning process or the factor screening optimization process, and updating the historical fault dataset on which the dynamic weight allocation depends.

8. A power transmission line monitoring and early warning system based on BeiDou communication, used to implement the power transmission line monitoring and early warning method based on BeiDou communication as described in claims 1-7, characterized in that, The system includes a spatial data acquisition and span modeling module, a multimodal factor screening and feature fusion module, an intelligent prediction and dynamic evaluation module, and a closed-loop collaboration and optimization module. The spatial data acquisition and span modeling module includes a BeiDou high-precision positioning submodule and a span unit modeling submodule. The BeiDou high-precision positioning submodule uses differential correction technology to calibrate tower coordinates, acquires data on transmission line routing, tower distribution, and topography, and extracts the boundaries of adjacent tower segments through ArcGIS to generate attribute information. Span unit modeling submodule: Integrates spatial-attribute information to generate a GeoJSON format span unit model, and filters key monitoring span units with overlapping fault areas through spatial overlay analysis; The multimodal factor screening and feature fusion module includes a rasterization factor processing submodule and a multimodal feature extraction submodule; Rasterization factor processing submodule: Divides the span unit into a grid, filters time-series dynamic factors, meteorological factors, line parameter factors and environmental factors, uses the natural discontinuity method for quantification and classification, and removes redundant factors through Pearson coefficient. Multimodal feature extraction submodule: Extracts deep features, concatenates them with the temporal dynamic factor time series matrix and the quantized hierarchical factor vector to form a fusion feature tensor, which is used as the model input; The intelligent prediction and dynamic assessment module includes a Transformer time-series prediction submodule, a multimodal fusion classification submodule, and a dynamic weight allocation submodule. The Transformer time-series prediction submodule extracts temporal periodic features through a causal convolutional network, encodes location information using a triangular temporal localization network, and predicts trends using an Encoder-Decoder architecture, outputting a predicted sequence. The multimodal fusion classification submodule concatenates time-series prediction features, image CNN features, and quantized grading factors, inputs them into a random forest classifier, outputs a fault tendency classification, and maps it to a fault tendency index. The dynamic weight allocation submodule optimizes the weight allocation of the fault probability index, fault impact index, and fault tendency index based on seasonal and real-time weather conditions using gradient descent, and combines this with the natural discontinuity method to divide the warning levels into four levels, clearly defining the response timeframe. The closed-loop collaboration and optimization module includes a warning information distribution submodule and a closed-loop management platform submodule; The early warning information distribution submodule distributes time-series early warning information, including the early warning level, grade unit number, risk trigger, dynamic weight basis and handling measures, and synchronizes it to the cloud platform to generate a time-series ledger; the closed-loop management platform submodule provides early warning response status feedback, updates on on-site verification progress, links handling results and generates closed-loop archives, and triggers model tuning or factor screening optimization through early warning consistency review.

9. A power transmission line monitoring and early warning device based on BeiDou communication, used to implement the power transmission line monitoring and early warning method based on BeiDou communication as described in claims 1-7, characterized in that, The device includes: Beidou high-precision positioning components: Beidou receivers integrating differential correction technology, tower coordinate calibration, and extraction of adjacent tower line segment boundaries via ArcGIS; Span unit modeling component: Based on compliance verification, it integrates spatial-attribute information to generate a GeoJSON format span unit model, filters key monitoring span units with overlapping fault areas through spatial overlay analysis, and generates attribute information by associating with power equipment management system data; Rasterization factor processor: Divides the span unit into a grid, filters time-series dynamic factors, meteorological factors, line parameter factors and environmental factors, uses the natural discontinuity method for quantification and classification, and eliminates redundant factors through Pearson coefficient. Multimodal feature extractor: Extracts depth features from UAV imagery and meteorological satellite cloud images, and concatenates them with the time series matrix of temporal dynamic factors and the quantized hierarchical factor vector to form a fusion feature tensor, which is used as the model input; Transformer temporal prediction component: Extracts temporal periodic features, combines them with triangular temporal localization network to encode location information, predicts trends, and outputs predicted sequences; Multimodal fusion classifier: It concatenates time-series prediction features, image CNN features and quantization grading factors, inputs them into a random forest classifier, outputs fault tendency classification, and maps it to a fault tendency index; Dynamic weight allocator: Based on seasonal and real-time weather conditions, the weight allocation of the fault probability index, fault impact index, and fault tendency index is optimized by gradient descent method, and the four-level early warning level is divided by natural discontinuity method, and the handling time limit is clearly defined. Early warning information distribution component: Distributes time-series early warning information, including early warning level, grade unit number, risk cause, dynamic weight basis and handling measures, and synchronizes it to the cloud platform to generate a time-series ledger; Closed-loop management platform components: An operation and maintenance platform integrating BeiDou communication interconnection to realize early warning response status feedback, on-site verification progress updates, handling result correlation and closed-loop file generation, and trigger model tuning or factor screening optimization through early warning consistency review.

Images