Power transmission line icing risk prediction method, system and equipment based on terrain and time sequence, and medium

Abstract

The invention discloses a terrain and time sequence-based power transmission line icing risk prediction method, system and device, and a medium. The prediction method comprises the following steps: acquiring meteorological elements, topographic features and time sequence features, unifying the meteorological elements, the topographic features and the time sequence features to the same time step length and geographic coordinate grid, and performing topographic correction on the meteorological elements according to the topographic features; constructing and optimizing an icing risk prediction model based on the meteorological elements after terrain correction and the time sequence features as input; inputting real-time weather forecast data into the optimized icing risk prediction model, dividing icing risk grades according to model output, and performing spatial visualization and early warning through a GIS platform; and quantizing the contribution of each element to a prediction result, optimizing model parameters and a feature extraction method, and carrying out interpretable driven model iteration. According to the invention, refined prediction and dynamic early warning of the icing risk are realized, accurate operation and maintenance of a power system are effectively supported, and the icing disaster risk is reduced.

Description

[0001] Technology Leader

[0002] The present invention belongs to the intersection of transmission line safety monitoring and artificial intelligence technology, and specifically relates to a method, system, equipment and medium for predicting the icing risk of transmission lines based on terrain and time series. Background Art

[0003] Icing of transmission lines is a major threat to power systems in cold regions. Traditional prediction methods rely on single meteorological models or empirical formulas, which fail to accurately reflect the impact of complex terrain and dynamic meteorological changes. These methods suffer from the following main drawbacks: 1. Single data dimension: ignoring the modulating effect of terrain undulations on the local microclimate; 2. Lack of temporal characteristics: ignoring the lag effects and periodic patterns of meteorological factors; 3. Model uninterpretability: unable to quantify the contribution of various factors to icing, resulting in a lack of targeted early warning measures. To address these shortcomings, a highly interpretable icing risk prediction method is urgently needed that integrates terrain characteristics and temporal dynamic information. Summary of the Invention

[0004] To address the shortcomings of existing methods, this invention provides a method, system, device, and medium for predicting transmission line icing risk based on terrain and time series. By integrating meteorological factors, terrain characteristics, and time series features, combined with machine learning models and interpretable analysis techniques, this invention enables refined prediction and dynamic early warning of icing risk.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] In a first aspect, a method for predicting the risk of icing on a transmission line based on terrain and time series includes the following steps:

[0007] S1. Acquire meteorological elements, terrain features, and time series features, unify them to the same time step and geographic coordinate grid, and perform terrain correction on the meteorological elements based on the terrain features;

[0008] S2. constructing and optimizing an icing risk prediction model based on the meteorological elements after terrain correction described in S1 and the time series characteristics as input;

[0009] S3. Input real-time weather forecast data into the optimized icing risk prediction model, classify icing risk levels based on model output, and perform spatial visualization and early warning through the GIS platform;

[0010] S4. Quantify the contribution of each factor to the prediction results, optimize model parameters and feature extraction methods, and perform interpretability-driven model iteration.

[0011] Preferably, the meteorological elements described in step S1 are extracted from multi-source meteorological data, and the multi-source meteorological data include temperature, air pressure, humidity, wind speed, wind direction, radiation elements and measured ice thickness provided by the PLAM model, WRF model and ground observation stations, and the multi-source meteorological data also includes remote sensing / reanalysis data.

[0012] Preferably, the terrain features in step S1 include slope, aspect and terrain roughness:

[0013]

[0014] Among them, Slope represents the slope, Aspect represents the slope direction, Δh represents the elevation change, and Δd represents the horizontal distance; and Represents the gradient of terrain data in the x and y directions respectively.

[0015] Preferably, the time series features described in step S1 include sliding window features, hysteresis features, and periodic features, and the sliding window features include:

[0016]

[0017] Among them, X mean,t is the sliding window mean of the time series at time t, X std,t is the sliding window standard deviation of the time series at time t, w is the size of the sliding window, X i is the original data value at the i-th time point in the time series, and t is the current time;

[0018] The hysteresis features include: multi-step lag processing of important meteorological elements to capture the possible delayed effects of icing;

[0019] The periodic characteristics include: if there are diurnal or seasonal changes, the time series can be subjected to discrete Fourier transform or wavelet transform to extract the periodic component.

[0020] Preferably, the terrain correction in step S1 includes correcting the temperature and wind speed based on the altitude, and correcting the radiation element based on the slope aspect, including:

[0021] The calculation formula for the temperature correction is:

[0022] T(h)=T(h0)-Γ·(h-h0)

[0023] Where Γ is the vertical lapse rate, h is the current altitude, and h0 is the reference altitude;

[0024] The calculation formula for the wind speed correction is:

[0025] Among them, W adj is the corrected wind speed, W raw is the original wind speed, α is the terrain roughness related index;

[0026] The radiation factor correction calculates the local incident radiation based on the angle between the slope aspect and the solar azimuth, and generates spatiotemporal interaction characteristics, including the product term of terrain roughness and wind speed within the same grid, the linear combination term of altitude and temperature, and the weighted sliding average term of historical ice thickness and current humidity.

[0027] Preferably, the ice risk prediction model construction method described in step S2 is a random forest RF or gradient boosting decision tree GBDT type algorithm; the ice risk prediction model optimization described in step S2 includes:

[0028] Parameter tuning is performed in the hyperparameter space using Grid Search or Bayesian Optimization algorithms, combined with k-fold cross-validation to evaluate the model's generalization ability and robustness, with k ≥ 5.

[0029] The output of the model includes ice thickness, ice probability and classification threshold. For ice thickness prediction, mean square error (MSE), mean absolute error (MAE) and determination coefficient (R) are used. 2 As evaluation indicators; for ice probability prediction, AUC value, F1-score, accuracy and Brier Score are selected for performance verification; for classification threshold, it is determined based on the dynamic cutoff point of the receiver operating characteristic curve.

[0030] Preferably, the step S3 of inputting the real-time weather forecast data into the trained model, classifying the ice risk level according to the model output, and performing spatial visualization and early warning through the GIS platform includes:

[0031] The icing probability P output by the model is mapped into four risk levels: P < 0.3, no risk; 0.3 ≤ P < 0.6, mild risk; 0.6 ≤ P < 0.8, moderate risk; P ≥ 0.8, severe risk; the terrain base map is superimposed on the GIS platform, and the spatial distribution of risk levels is displayed in the form of a heat map. The warning signal trigger threshold is generated, and manual inspection is initiated when P ≥ 0.6.

[0032] Preferably, the step S4 of quantifying the contribution of each factor to the prediction result, optimizing the model parameters and feature extraction method, and performing interpretability-driven model iteration includes:

[0033] The contribution of each feature in a single sample is quantified based on the SHAP value, and the features ranked in the top 20% in terms of contribution are selected as core factors. When the absolute error between the model prediction result and the measured ice thickness is ≥30%, the feedback mechanism is automatically triggered to retrain the model and optimize the terrain correction parameters, and a visual report is generated based on the SHAP value analysis.

[0034] In a second aspect, the present invention provides a system for predicting the risk of icing on power transmission lines based on terrain and time series, comprising:

[0035] A data acquisition module is used to acquire meteorological elements, terrain features and time series features, unify them to the same time step and geographic coordinate grid, and perform terrain correction on the meteorological elements based on the terrain features;

[0036] A model building and optimization module, configured to build and optimize an icing risk prediction model based on the terrain-corrected meteorological elements and the time series characteristics as input;

[0037] The prediction module is used to input real-time weather forecast data into the optimized icing risk prediction model, classify icing risk levels based on the model output, and perform spatial visualization and early warning through the GIS platform;

[0038] The explanation module is used to quantify the contribution of each factor to the prediction results, optimize model parameters and feature extraction methods, and perform interpretability-driven model iteration.

[0039] In a third aspect, the present invention provides an electronic device comprising a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement any one of the above-mentioned prediction methods.

[0040] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed by a processor, any one of the above-mentioned prediction methods is implemented.

[0041] The positive beneficial effects of the present invention are:

[0042] 1. By integrating meteorological elements, terrain features and time series characteristics and combining terrain correction technology, this invention effectively solves the problem of traditional methods ignoring the terrain microclimate effect, and significantly improves the temporal and spatial resolution and accuracy of icing risk prediction.

[0043] 2. This invention incorporates time series features to capture the lag effects and seasonal patterns of meteorological factors, addressing the inadequacy of existing technologies in modeling dynamic time series features. By combining machine learning models with grid search or Bayesian optimization, it achieves nonlinear modeling of icing risk, significantly reducing the mean square error of ice thickness prediction.

[0044] 3. This invention uses SHAP values ​​to quantify the contribution of meteorological factors, terrain characteristics, and time series characteristics to icing risk, solving the unexplainable problem of the traditional "black box model." By dynamically adjusting the classification threshold (ROC curve cutoff point) and risk level division (four-level warning), combined with GIS platform visualization, the warning triggering accuracy rate is increased to 80.9% (icing category), supporting the power department to accurately deploy inspection resources.

[0045] 4. This invention uses real-time weather forecast data (WRF and PLAM models updated hourly) to dynamically predict icing risk. A feedback mechanism optimizes model parameters (e.g., retraining when errors exceed 30%), ensuring robustness in complex terrain and extreme weather conditions. Practical case studies have demonstrated that the model's overall accuracy reaches 87.6% in high-icing areas such as the mountainous areas of Henan Province, while the recall rate for non-icing events increases to 94.8%, significantly reducing power system operation and maintenance costs.

[0046] 5. The present invention integrates multi-source meteorological data, terrain features, and time series information, combines terrain correction technology and time series features, and constructs a machine learning model based on random forest or gradient boosting decision trees. Bayesian optimization is used to improve generalization capabilities. The model outputs ice thickness or risk probability, dynamically predicts risk levels based on real-time meteorological forecast data, utilizes a GIS platform for visual warnings, triggers manual inspections, and implements interpretable analysis by quantifying feature contributions. The present invention achieves refined prediction and dynamic warning of ice risk, solving the technical problems of insufficient prediction accuracy caused by existing ice prediction methods, which neglect terrain microclimate effects and lack time series dynamic features. It effectively supports the precise operation and maintenance of power systems and reduces the risk of ice disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A fuzzy matrix diagram of the ice cover prediction results provided by an embodiment of the present invention.

[0048] Figure 2 An interpretable diagram of the ice cover prediction results provided by an embodiment of the present invention.

[0049] Figure 3 An interpretable SHAP diagram of the importance of various meteorological factors to ice cover prediction provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The present invention is described in detail below with reference to the accompanying drawings to make the technical route and operation steps of the present invention clearer.

[0051] Example 1

[0052] A method for predicting the icing risk of power transmission lines based on terrain and time series includes the following steps:

[0053] S1. Acquire meteorological elements, terrain features, and time series features, and unify them to the same time step Δt (Δt ≤ 1 hour) and geographic coordinate grid (resolution ≤ 1 km × 1 km) using a space-time matching algorithm, wherein the space-time matching algorithm includes bilinear interpolation and kriging interpolation; perform terrain correction on the meteorological elements based on the terrain features;

[0054] Furthermore, meteorological elements are extracted from multi-source meteorological data, including temperature, air pressure, humidity, wind speed, wind direction, radiation elements and measured ice thickness provided by the PLAM model, WRF model and ground observation stations. Multi-source meteorological data also includes remote sensing / reanalysis data: such as ERA5 reanalysis data, which is used to supplement the meteorological field with a large range of temporal and spatial resolution.

[0055] Furthermore, terrain features are extracted from terrain data, including: obtaining terrain height information of the study area, with the resolution selected according to the research accuracy requirements; slope and aspect are derived and calculated from terrain data, reflecting the inclination and direction of the local slope; terrain roughness: for example, terrain undulation and cutting depth indicators are used to measure complex terrain environments:

[0056]

[0057] Among them, Slope represents the slope, Aspect represents the slope direction, Δh represents the elevation change, and Δd represents the horizontal distance; and Represents the gradient of terrain data in the x and y directions respectively.

[0058] Furthermore, time series features are extracted from the time series information, including sliding window features, lag features, and periodic features. The sliding window features include:

[0059] By setting the window size w, the mean, standard deviation, and extreme value features are extracted to characterize the changing trend of meteorological elements in the recent time period:

[0060]

[0061] Among them, X mean,t is the sliding window mean of the time series at time t, Xstd,t is the sliding window standard deviation of the time series at time t, w is the size of the sliding window (unit: time step), X i is the original data value at the i-th time point in the time series, and t is the current time;

[0062] Hysteresis features include: multi-step lag processing of important meteorological elements (such as temperature, humidity, and precipitation) to capture the possible delayed effects of icing;

[0063] Periodic characteristics include: If there are diurnal or seasonal variations, the time series can be subjected to discrete Fourier transform (DFT) or wavelet transform to extract the periodic components.

[0064] Furthermore, terrain correction includes altitude corrections for temperature and wind speed based on vertical lapse rate, and corrections for radiation elements based on aspect:

[0065] When correcting from the reference altitude h0 to the altitude h, the temperature and wind speed are corrected linearly or empirically. The temperature correction is calculated as follows:

[0066] T(h)=T(h0)-Γ·(h-h0)

[0067] Where Γ is the vertical lapse rate (0.006-0.008 K / m), h is the current altitude, and h0 is the reference altitude;

[0068] The calculation formula for wind speed correction is:

[0069] Among them, W adj is the corrected wind speed, W raw is the original wind speed, α is the terrain roughness related index, and its value ranges from 0.1 to 0.3;

[0070] The radiation factor correction calculates the local incident radiation based on the angle between the slope aspect and the solar azimuth, and generates spatiotemporal interaction characteristics, including the product term of terrain roughness and wind speed within the same grid, the linear combination term of altitude and temperature, and the weighted sliding average term of historical ice thickness and current humidity.

[0071] S2. constructing and optimizing an icing risk prediction model based on the terrain-corrected meteorological elements and the time series characteristics as input;

[0072] Model selection: Random forest (RF) or gradient boosting decision tree (GBDT) algorithms (such as LightGBM, XGBoost, and CatBoost) are used to construct an icing risk prediction model. The gradient boosting decision tree (GBDT) uses the LightGBM framework and enables histogram acceleration and automatic encoding of category features. During training, a weight coefficient of ω = 2 to 5 is applied to icing samples to alleviate category imbalance.

[0073] Model training: Grid search or Bayesian optimization algorithms are used to search for the optimal combination of hyperparameters, such as the tree depth (max_depth) and the minimum number of samples for splitting leaf nodes (min_child_samples). The model's generalization ability is evaluated based on k-fold cross-validation (k ≥ 5). In other words, k-fold cross-validation can be used during training to reduce overfitting and evaluate model robustness.

[0074] Model evaluation metrics: For regression scenarios (ice thickness), mean square error (MSE), mean absolute error (MAE), and coefficient of determination (R^2) are used for evaluation; for classification or probability scenarios (ice cover probability), AUC (area under the receiver operating characteristic (ROC) curve), F1-score, accuracy, and Brier Score are used for evaluation.

[0075] Model outputs: ice thickness (regression task), ice probability (classification task), and classification threshold. The classification threshold is determined by the dynamic cutoff point of the receiver operating characteristic curve (ROC).

[0076] S3. Input real-time weather forecast data into the optimized icing risk prediction model, classify icing risk levels based on model output, and perform spatial visualization and early warning through the GIS platform;

[0077] Real-time / quasi-real-time prediction: The regularly updated weather forecast from the Numerical Weather Forecasting (WRF) model is fed into a trained model. Combined with current terrain and time series information, the model outputs ice thickness or risk probability at fixed intervals (e.g., 1 hour or less). The results are visualized on a GIS platform or monitoring system, showing spatial distribution and forecast results for future periods.

[0078] Risk warning and decision support: Based on the ice thickness or ice probability output by the model, several risk levels (e.g., no risk, mild, moderate, severe) are set and marked with different colors on the map. Early warnings or inspections are carried out in coordination with the power operation and maintenance department's strategies.

[0079] The icing probability P output by the model is mapped into four risk levels: no risk (P < 0.3), mild risk (0.3 ≤ P < 0.6), moderate risk (0.6 ≤ P < 0.8), and severe risk (P ≥ 0.8). A topographic base map is overlaid on the GIS platform to display the spatial distribution of risk levels in the form of a heat map, and an early warning signal trigger threshold is generated (manual inspection is initiated when P ≥ 0.6). The GIS platform integrates a weather forecast update interface and receives forecast data from the WRF model every Δt time interval.

[0080] S4. Quantify the contribution of each factor to the prediction results, optimize model parameters and feature extraction methods, and perform interpretability-driven model iteration;

[0081] Feature Contribution Analysis: Quantify the contribution of meteorological, terrain, and time series features to icing risk prediction using the feature importance (Feature Importance) or Shapley Additive Explanations (SHAP) values ​​based on tree models, focusing on the relative impact of core elements such as terrain height / aspect, wind speed, temperature, and humidity.

[0082] The SHAP value is based on the concept of Shapley Value in game theory. The contribution of a single feature j to the prediction result can be written as:

[0083]

[0084] Among them, φ j is the contribution of a single feature j to the prediction result, f(S∪{j}) is the model output, M is the total number of features, and S is the feature subset;

[0085] Feedback and model improvement: Compare actual observed ice cover with model predictions, update training samples or readjust parameters, and continuously improve terrain correction methods or time series feature extraction methods.

[0086] The contribution of each feature in a single sample is quantified based on the SHAP value, and the features ranked in the top 20% in terms of contribution are selected as core factors. When the absolute error between the model prediction result and the measured ice thickness is ≥30%, the feedback mechanism is automatically triggered to retrain the model and optimize the terrain correction parameters. Two types of visualization reports are generated based on the SHAP value analysis: a global explanation report, which includes the average contribution ranking of temperature, humidity, wind speed, precipitation and terrain roughness; and a case explanation report, which includes the annotation of the dominant factors and their contribution values ​​for high-risk points (P ≥ 0.8).

[0087] Application test

[0088] The method for predicting icing risk of power transmission lines based on terrain and time series provided by the present invention was verified and applied in the mountainous area of ​​Xinyang, Henan.

[0089] Historical ice cover data and multi-source meteorological data from several key micro-topographic areas in the Henan power grid system (including Taohuazhai, Sanmenxia Xiaoshan, Nanyang Fangshan, Zhengzhou Jianshan, and Shizushan) were selected to verify and predict the random forest (RF) model. By comparing the model prediction results with the actual ice cover conditions, the prediction ability of the model under different meteorological conditions was evaluated.

[0090] For the "none-ice" category, the model achieved a high accuracy of 0.89, a recall of 0.95, and an F1 score of 0.92, demonstrating its strong predictive power for non-ice events. This high recall indicates that the model is able to effectively identify the majority of periods without ice, particularly during mild weather conditions, accurately predicting these periods as non-ice events.

[0091] For the ice category, although the model has a precision of 0.81, its recall is low, only 0.66, and its F1 score is 0.72. This indicates that the model's prediction accuracy for ice-covered periods is low, especially in the face of extreme weather and snow events. The model has difficulty accurately identifying all ice-covered events. This may be due to the relatively small number of ice-covered phenomena and the small number of ice-covered samples in the dataset, which leads to certain deviations in the model training.

[0092] The comprehensive evaluation results show that the overall accuracy of the model is 0.88, which means that among all predicted samples, about 88% of the samples are correctly classified. This is a very high accuracy rate, indicating that the model has a strong classification ability for all samples as a whole. The macro-average precision (0.85) and recall rate (0.80) show that the performance of the model is relatively balanced across all categories. Although the recall rate for the "ice-covered" category is slightly lower, the model as a whole has no significant category bias. The weighted average precision (0.87) and recall rate (0.88) show that considering the weight of the category samples, the model has a good classification performance for most samples, especially for the ice-covered category.

[0093] Furthermore, in actual cases, considering that icing in Henan Province is primarily concentrated in areas with unique microtopography and microclimates, such as Sanmenxia and Xinyang, which are typically located in mountainous, hilly, or high-altitude areas, icing of transmission lines is common in these areas due to their complex topography and variable climate. To verify the effectiveness of the icing prediction model, this case study selected all samples as the overall case to conduct an interpretability analysis of the model. The results are discussed below:

[0094] Accuracy: The overall accuracy is (166+38) / 233≈87.6%, indicating that the model can correctly classify iced and non-iced areas in most cases.

[0095] Precision for Ice: The accuracy is 38 / (38+9)≈80.9%, indicating that when the model predicts a sample as ice-covered, there is an 80.9% probability that it is actually ice-covered.

[0096] Recall for Ice: The recall is 38 / (38+20)≈65.5%, indicating that the model can capture 65.5% of actual icing events.

[0097] Figure 1 is the fuzzy matrix diagram of ice cover prediction results, Figure 1 The displayed confusion matrix is ​​an important tool for evaluating the performance of classification models. It shows the relationship between the actual categories and the categories predicted by the model. The key information in the confusion matrix provided by the present invention includes:

[0098] Non-Ice Prediction: True Positives (TP): The model correctly predicted non-ice as non-ice 166 times. False Negatives (FN): The model incorrectly predicted 9 samples as iced when they were actually non-ice.

[0099] Ice Prediction: True Positives (TP): The model correctly predicted ice as ice 38 times. False Negatives (FN): The model incorrectly predicted 20 samples that were actually ice as non-ice.

[0100] Furthermore, confusion matrix analysis shows that the model performs well when predicting non-icing events, with high precision and recall. However, when predicting icing events, despite high precision, the recall is relatively low. This suggests that the model fails to fully capture all icing scenarios under specific climate and terrain conditions, especially in areas with drastic climate changes and complex terrain. Furthermore, prediction errors tend to increase under extreme meteorological conditions (such as severe cold waves and heavy precipitation). This is primarily due to the unpredictability of atmospheric convection and complex weather systems.

[0101] Figure 2This is an interpretability diagram of the ice prediction results, where class 0 represents un-iced areas and class 1 represents iced areas. The SHAP value shows the average contribution of each feature to the model prediction results. SHAP value is an advanced model interpretation method used to explain the specific contribution of each feature to the model prediction results. The key information in the interpretability diagram of the ice prediction results provided by the present invention includes:

[0102] The impact of features on prediction:

[0103] Temperature: For Class 1 (icing), the effect of temperature is significantly greater, indicating that lower temperatures are a key factor in icing and help the model predict icing conditions. For Class 0 (unicing), the effect of temperature is relatively small, indicating that temperature may not be the dominant factor in predicting unicing events.

[0104] Humidity: For Class 1, humidity contributes significantly to icing predictions, indicating that the model tends to predict icing in high-humidity environments. This is likely because moisture in the air under high-humidity conditions is more likely to form ice when cooled. For Class 0, humidity has a smaller impact, indicating that humidity plays a less significant role than temperature in predicting no icing.

[0105] Wind speed: Wind speed has a significant positive impact on Class 1, indicating that higher wind speeds are positively correlated with ice cover in the model. This is likely because high wind speeds help lower surface temperatures and accelerate ice formation. For Class 0, wind speed has a smaller impact, indicating that wind speed is not a decisive predictor in uniced conditions.

[0106] Precipitation: For Class 1, precipitation has a significant positive impact on the prediction of ice cover, indicating that precipitation is a strong indicator of ice cover events and provides the necessary moisture for ice formation. For Class 0, precipitation has a small impact, indicating that precipitation does not contribute much to the prediction of un-iced events.

[0107] Figure 3 This is an interpretable SHAP diagram of the importance of various meteorological factors to ice cover prediction. The key information in the interpretable SHAP diagram of the importance of various meteorological factors to ice cover prediction provided by the present invention includes:

[0108] Figure 3The SHAP values ​​of different meteorological factors (temperature, humidity, wind speed, and precipitation) in the icing prediction model are displayed. These values ​​reflect the magnitude and direction of each feature's influence on the model output. This visualization allows for detailed analysis of the behavior and contribution of each feature in the prediction process. The SHAP value distribution shows that the temperature feature has significant positive and negative effects, with a wide distribution at both the positive and negative ends of the SHAP value spectrum. This indicates that the influence of temperature on the model is complex and variable: lower temperatures may promote model predictions of icing events, while higher temperatures tend to predict no icing. The SHAP values ​​of humidity are primarily concentrated in the positive region and have a relatively small impact, indicating that in most cases, higher humidity tends to increase the probability of model predictions of icing. This is consistent with the actual influence of humidity on ice formation, especially under low temperature conditions. The SHAP values ​​of wind speed are widely distributed near zero, indicating that its influence on the model output is less pronounced than that of temperature and humidity. Lower wind speeds may slightly promote icing, while higher wind speeds may hinder icing formation because they may increase heat exchange and inhibit surface cooling. The SHAP values ​​for precipitation are concentrated in the positive region, indicating that precipitation has a consistent and significant impact on the model's icing predictions. This is consistent with actual conditions, as precipitation provides the moisture necessary for ice formation and is the direct cause of icing events. The results provided by SHAP analysis enable power grid maintenance teams to take more accurate preventative measures before severe winter weather. For example, by understanding the impact of precipitation and low temperatures on the probability of icing, teams can more effectively deploy emergency resources and reduce power outages caused by icing.

[0109] Furthermore, the model demonstrated high accuracy across most time periods, particularly in low-temperature and low-humidity conditions. For areas with low wind speeds (2-3 m / s) and near-freezing temperatures, the model was able to better capture icing risks.

[0110] Example 2

[0111] A system for predicting transmission line icing risk based on terrain and time series, used to implement the method described in Example 1, comprising:

[0112] A data acquisition module is used to acquire meteorological elements, terrain features and time series features, unify them to the same time step and geographic coordinate grid, and perform terrain correction on the meteorological elements based on the terrain features;

[0113] A model building and optimization module, configured to build and optimize an icing risk prediction model based on the terrain-corrected meteorological elements and the time series characteristics as input;

[0114] The prediction module is used to input real-time weather forecast data into the optimized icing risk prediction model, classify icing risk levels based on the model output, and perform spatial visualization and early warning through the GIS platform;

[0115] The explanation module is used to quantify the contribution of each factor to the prediction results, optimize model parameters and feature extraction methods, and perform interpretability-driven model iteration.

[0116] Example 3

[0117] Based on similar inventive concepts, the present invention provides an electronic device, including a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement the prediction method described in the above-mentioned embodiment 1.

[0118] The memory can be used to store computer programs, and the processor implements the method described in Example 1 by executing the computer programs stored in the memory and calling the data stored in the memory.

[0119] The memory may mainly include a program storage area and a data storage area. The program storage area may store an operating system, at least one application required for a function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device (such as audio data), etc. In addition, the memory may include non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device.

[0120] The processor can be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device and connects all parts of the electronic device using various interfaces and lines.

[0121] Example 4

[0122] Based on similar inventive concepts, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed by a processor, the prediction method described in the above embodiment 1 is implemented.

[0123] If the module / unit integrated in the electronic device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. Computer-readable media may include: any entity or device that can carry computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory and read-only memory (ROM, Read-Only Memory).

[0124] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiment of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal translation scripting language JavaScript, etc.

[0125] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0126] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0127] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0128] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the risk of icing on power transmission lines based on terrain and time series, characterized in that: The steps include: S1. Acquire meteorological elements, terrain features, and time series features, unify them to the same time step and geographic coordinate grid, and perform terrain correction on the meteorological elements based on the terrain features; S2. constructing and optimizing an icing risk prediction model based on the meteorological elements after terrain correction described in S1 and the time series characteristics as input; S3. Input real-time weather forecast data into the optimized icing risk prediction model, classify icing risk levels based on model output, and perform spatial visualization and early warning through the GIS platform; S4. Quantify the contribution of each factor to the prediction results, optimize model parameters and feature extraction methods, and perform interpretability-driven model iteration.

2. The method for predicting transmission line icing risk based on terrain and time series according to claim 1 is characterized in that: The meteorological elements described in step S1 are extracted from multi-source meteorological data, and the multi-source meteorological data include temperature, air pressure, humidity, wind speed, wind direction, radiation elements and measured ice thickness provided by the PLAM model, the WRF model and ground observation stations, and the multi-source meteorological data also includes remote sensing / reanalysis data.

3. The method for predicting transmission line icing risk based on terrain and time series according to claim 1, characterized in that: The terrain characteristics in step S1 include slope, aspect and terrain roughness: Among them, Slope represents the slope, Aspect represents the slope direction, Δh represents the elevation change, and Δd represents the horizontal distance; and Represents the gradient of terrain data in the x and y directions respectively.

4. The method for predicting transmission line icing risk based on terrain and time series according to claim 1, characterized in that: The time series features described in step S1 include sliding window features, hysteresis features, and periodic features. The sliding window features include: Among them, X mean,t is the sliding window mean of the time series at time t, X std,r is the sliding window standard deviation of the time series at time t, w is the size of the sliding window, X i is the original data value at the i-th time point in the time series, and t is the current time; The hysteresis features include: multi-step lag processing of important meteorological elements to capture the possible delayed effects of icing; The periodic characteristics include: if there are diurnal or seasonal changes, the time series can be subjected to discrete Fourier transform or wavelet transform to extract the periodic component.

5. The method for predicting transmission line icing risk based on terrain and time series according to claim 1, characterized in that: The terrain correction described in step S1 includes correcting temperature and wind speed based on altitude, and correcting radiation elements based on aspect, including: The calculation formula for the temperature correction is: T(h)=T(h0)-Γ·(h-h0) Where Γ is the vertical lapse rate, h is the current altitude, and h0 is the reference altitude; The calculation formula for the wind speed correction is: Among them, W adj is the corrected wind speed, W raw is the original wind speed, α is the terrain roughness related index; The radiation factor correction calculates the local incident radiation based on the angle between the slope aspect and the solar azimuth, and generates spatiotemporal interaction characteristics, including the product term of terrain roughness and wind speed within the same grid, the linear combination term of altitude and temperature, and the weighted sliding average term of historical ice thickness and current humidity.

6. The method for predicting transmission line icing risk based on terrain and time series according to claim 1, characterized in that: The ice risk prediction model construction method described in step S2 is a random forest RF or gradient boosting decision tree GBDT type algorithm; the ice risk prediction model optimization described in step S2 includes: Parameter tuning is performed in the hyperparameter space using Grid Search or Bayesian Optimization algorithms, combined with k-fold cross-validation to evaluate the model's generalization ability and robustness, with k ≥ 5. The output of the model includes ice thickness, ice probability and classification threshold. For ice thickness prediction, mean square error (MSE), mean absolute error (MAE) and determination coefficient (R) are used. 2 As evaluation indicators; for ice probability prediction, AUC value, F1-score, accuracy and Brier Score are selected for performance verification; for classification threshold, it is determined based on the dynamic cutoff point of the receiver operating characteristic curve.

7. The method for predicting transmission line icing risk based on terrain and time series according to claim 1, characterized in that: Step S3 includes inputting real-time weather forecast data into the trained model, classifying ice risk levels based on the model output, and performing spatial visualization and early warning through the GIS platform, including: The icing probability P output by the model is mapped into four risk levels: P < 0.3, no risk; 0.3 ≤ P < 0.6, mild risk; 0.6 ≤ P < 0.8, moderate risk; P ≥ 0.8, severe risk; the terrain base map is superimposed on the GIS platform, and the spatial distribution of risk levels is displayed in the form of a heat map. The warning signal trigger threshold is generated, and manual inspection is initiated when P ≥ 0.

6.

8. The method for predicting transmission line icing risk based on terrain and time series according to claim 1, characterized in that: Step S4, which quantifies the contribution of each factor to the prediction result, optimizes model parameters and feature extraction methods, and performs interpretability-driven model iteration, includes: The contribution of each feature in a single sample is quantified based on the SHAP value, and the features ranked in the top 20% in terms of contribution are selected as core factors. When the absolute error between the model prediction result and the measured ice thickness is ≥30%, the feedback mechanism is automatically triggered to retrain the model and optimize the terrain correction parameters, and a visual report is generated based on the SHAP value analysis.

9. A prediction system for transmission line icing risk based on terrain and time series, characterized by: include: A data acquisition module is used to acquire meteorological elements, terrain features and time series features, unify them to the same time step and geographic coordinate grid, and perform terrain correction on the meteorological elements based on the terrain features; A model building and optimization module, configured to build and optimize an icing risk prediction model based on the terrain-corrected meteorological elements and the time series characteristics as input; The prediction module is used to input real-time weather forecast data into the optimized icing risk prediction model, classify icing risk levels based on the model output, and perform spatial visualization and early warning through the GIS platform; The explanation module is used to quantify the contribution of each factor to the prediction results, optimize model parameters and feature extraction methods, and perform interpretability-driven model iteration.

10. An electronic device, characterized in that: The system comprises a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement the prediction method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed by a processor, the prediction method according to any one of claims 1 to 8 is implemented.

Images