Mountainous area power grid fire hazard prediction method and system

Abstract

The invention relates to the technical field of power system safety engineering and natural disaster risk assessment, in particular to a mountain area power grid fire prediction method and system, and the method comprises the steps: carrying out the grid division of a region where a mountain area power grid is located; extracting a multi-dimensional data feature corresponding to each grid; inputting the multi-dimensional data features into a pre-trained XGBoost machine learning model to obtain a fire risk value of the region, and mapping the fire risk value of the region and a corresponding mountain power grid to obtain a fire risk value of the mountain power grid; wherein the scale of the network is not greater than 10m * 10m; the multi-dimensional data features comprise topographic features, vegetation features and meteorological features; according to the method, multi-source data can be fused, fine risk division of a ten-meter spatial scale is realized, and the method has a dynamic risk quantification capability, so that a reliable decision basis is provided for forest fire prevention and control.

Description

Technical Field

[0001] This invention relates to the fields of power system safety engineering and natural disaster risk assessment technology, specifically to a method and system for predicting fires in mountainous power grids. Background Technology

[0002] Mountain power grids are crucial for power transmission, but their corridors often traverse towering mountains, facing a severe threat of wildfires. Once a wildfire breaks out, it can easily cause line tripping, equipment damage, and widespread power outages, posing a significant risk to the safe and stable operation of the power grid. Currently, existing wildfire risk assessment methods have significant shortcomings in achieving "refined zoning" and "quantitative prediction," specifically as follows: (1) The spatial resolution of the zoning is insufficient. The resolution of traditional zoning methods is usually at the kilometer level. In mountainous areas, the terrain is often undulating and the vegetation distribution is highly varied. It is difficult to capture the risk differences caused by terrain changes with coarse resolution. At the same time, the coarse scale will lead to the homogenization of risk "hot spots", and the generated zoning map cannot provide accurate guidance for the risk prevention and control of transmission lines.

[0003] (2) Existing methods rely on a few data sources and fail to effectively combine geographical environmental factors and combustible factors related to the spatiotemporal distribution of wildfire lines. This lack of data dimensions will cause the risk index to fail to effectively reflect the real situation and make it difficult to achieve accurate prediction.

[0004] (3) Existing technologies are mostly based on static, qualitative or semi-quantitative assessments, lacking the ability to make real-time dynamic quantitative predictions. They cannot provide risk probability values ​​at specific spatial locations, and the prediction information is relatively vague, which cannot provide sufficient assistance to the operation and maintenance department in formulating air defense measures.

[0005] Therefore, there is an urgent need to design a more accurate method for predicting power grid fires in mountainous areas. Summary of the Invention

[0006] To address the aforementioned problems in existing technologies, this invention proposes a method for predicting power grid fires in mountainous areas, comprising: Grid division is carried out for the area where the power grid is located in the mountainous region; Extract the multidimensional data features corresponding to each grid; The multidimensional data features are input into a pre-trained XGBoost machine learning model to obtain the fire risk value of the area, and the fire risk value of the area is mapped to the corresponding mountain power grid to obtain the fire risk value of the mountain power grid. The network scale is no larger than 10m×10m; the multidimensional data features include terrain features, vegetation features and meteorological features.

[0007] Optionally, the training process of the XGBoost machine learning model is as follows: The XGBoost machine learning model is trained using the grid where the historical thermal power plant is located and the corresponding multidimensional data features as positive samples, and the grid where the non-ignition location is located and the corresponding multidimensional data features as negative samples.

[0008] Optionally, the mountain power grid fire prediction method further includes: The fire risk level is obtained by comparing the fire risk value with a preset classification threshold. Based on the fire risk level, the corresponding mountain power grid fire risk value is mapped and an early warning is issued.

[0009] Optionally, after obtaining the fire risk level, it may also include visualizing the risk level on an electronic map using different colors.

[0010] Optionally, the terrain features are obtained by extracting terrain data, which includes at least one of elevation, slope, and aspect. The meteorological characteristics are obtained by extracting meteorological data, which includes at least one of surface temperature, soil moisture, cumulative precipitation, and wind direction and speed. The vegetation features are obtained by extracting vegetation data, which includes at least one of tree species, combustible load, and leaf index, wherein the combustible load is obtained by tree species inversion.

[0011] Optionally, the mountain power grid fire prediction method further includes resampling the meteorological data, terrain data, and vegetation data to a 10-meter resolution.

[0012] Optionally, a bilinear interpolation algorithm can be used for resampling.

[0013] Secondly, the present invention also provides a mountain power grid fire prediction system, comprising: The grid division module is used to divide the area where the power grid is located in mountainous areas into grids; The feature extraction module is used to extract multidimensional data features corresponding to each grid. The calculation module is used to input the multidimensional data features into a pre-trained XGBoost machine learning model to obtain the fire risk value of the area, and to map the fire risk value of the area to the corresponding mountain power grid to obtain the fire risk value of the mountain power grid. The network scale is no larger than 10m×10m; the multidimensional data features include terrain features, vegetation features and meteorological features.

[0014] Optionally, the training process of the XGBoost machine learning model in the computing module is as follows: The XGBoost machine learning model is trained using the grid where the historical thermal power plant is located and the corresponding multidimensional data features as positive samples, and the grid where the non-ignition location is located and the corresponding multidimensional data features as negative samples.

[0015] Optionally, the mountain power grid fire prediction system further includes a classification module, which is used for: The fire risk level is obtained by comparing the fire risk value with a preset classification threshold. Based on the fire risk level, the corresponding mountain power grid fire risk value is mapped and an early warning is issued.

[0016] Optionally, after obtaining the fire risk level, the risk classification module may also visualize the risk level on an electronic map using different colors.

[0017] Optionally, the terrain features in the feature extraction module are obtained by extracting terrain data, which includes at least one of elevation, slope, and aspect. The meteorological characteristics are obtained by extracting meteorological data, which includes at least one of surface temperature, soil moisture, cumulative precipitation, and wind direction and speed. The vegetation features are obtained by extracting vegetation data, which includes at least one of tree species, combustible load, and leaf index, wherein the combustible load is obtained by tree species inversion.

[0018] Optionally, the mountain power grid fire prediction method further includes a preprocessing module for resampling the meteorological data, terrain data, and vegetation data to a 10-meter resolution.

[0019] Optionally, the preprocessing module uses a bilinear interpolation algorithm for resampling.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention provides a method and system for predicting fires in mountainous power grids, including dividing the area where the power grid is located into grids; extracting multidimensional data features corresponding to each grid; inputting the multidimensional data features into a pre-trained XGBoost machine learning model to obtain the fire risk value of the area; and mapping the fire risk value of the area to the corresponding mountainous power grid to obtain the fire risk value of the mountainous power grid. The grid scale is no larger than 10m × 10m. The multidimensional data features include terrain features, vegetation features, and meteorological features. This invention, for the first time, improves the spatial accuracy of mountainous power grid fire risk assessment from the traditional "kilometer level" to the "ten-meter level," and achieves accurate risk mapping from geographical grids to specific power system line sections. This invention also creates an XGBoost risk quantification model that integrates static environmental and dynamic meteorological factors. This invention is the first to deeply and nonlinearly fuse static environmental factors such as terrain and vegetation with dynamic meteorological factors using the XGBoost algorithm, and outputs a continuous and comparable risk score, realizing the transformation of risk from qualitative judgment to quantitative description. The final output of this invention is not simply a risk map, but a "section-level early warning list" that can directly guide power grid operation and maintenance, combining the algorithm with the actual application of power. Attached Figure Description

[0021] Figure 1 This is a flowchart of the mountain power grid fire prediction method proposed in this invention; Figure 2 The present invention proposes Figure 1 A schematic diagram of the further process after step S3 in the Zhongshan District power grid fire prediction method; Figure 3 This is the application process flow of the mountain power grid fire prediction method proposed in this invention; Figure 4 This is a schematic diagram of the overall process of the mountain power grid fire prediction method proposed in this invention; Figure 5 This is a schematic diagram of the electronic device proposed in this invention. Detailed Implementation

[0022] This invention proposes a method and system for predicting fires in mountainous power grids. It is a method for refined zoning and quantitative prediction of fire risks in mountainous power grids that combines geographic information systems (GIS), remote sensing technology, meteorology, and data mining. It is particularly suitable for wildfire early warning, precise allocation of operation and maintenance resources, and disaster prevention and mitigation decision support in transmission line corridors with complex terrain.

[0023] Example 1: A method for predicting power grid fires in mountainous areas, such as Figure 1 As shown, it includes the following steps S1 to S3.

[0024] S1, divide the area where the power grid is located in the mountainous area into grids; the scale of the grid is no larger than 10m×10m.

[0025] Collect data closely related to wildfire occurrences, including topographic data, meteorological data, and vegetation data.

[0026] Topographic data, including elevation, slope, and aspect, can be obtained through satellite or open-source data to acquire a digital elevation model (DEM) of the target area.

[0027] Meteorological data, including surface temperature, soil moisture, cumulative precipitation, wind direction and speed, can be obtained from ground weather stations or reanalysis data such as ERA5.

[0028] Vegetation data includes tree species, combustible load, and leaf area index. It mainly utilizes high-resolution satellite imagery, using its multispectral bands and a random forest classification algorithm to identify tree species (broadleaf forest, shrubland, coniferous forest, etc.) in each pixel, and performs operations to retrieve combustible load, leaf area index, etc.

[0029] Within the GIS platform, the collected multi-source data is unified into a geographic coordinate system and projection system. Data with different resolutions are resampled to 10-meter resolution using algorithms such as bilinear interpolation, and then normalized to eliminate the influence of dimensions.

[0030] Using ArcGIS or QGIS software, with the power transmission and distribution channels of the mountain power grid as the core area, a 10m×10m regular grid covering the target area is generated, and the evaluation area is discretized into several independent evaluation units.

[0031] S2, extract the multidimensional data features corresponding to each grid; the multidimensional data features include terrain features, vegetation features and meteorological features.

[0032] This step aims to transform the preprocessed multi-source data into structured features that can be recognized and learned by machine learning models. Its core process involves extracting or calculating one or more feature values ​​from each corresponding data layer for each grid cell no larger than 10m × 10m, ultimately combining all features into a multi-dimensional feature vector. The extracted features can be categorized as follows: (1) Topographic features Topographic features describe the geographical conditions of the grid location. By processing digital elevation model data, the slope, elevation, and aspect of each grid are extracted. These factors together affect the speed and direction of wildfire spread.

[0033] (2) Vegetation characteristics Vegetation characteristics describe the combustible material situation within the grid. Satellite imagery is used to match specific vegetation types for each grid (such as easily ignitable coniferous forests), and vegetation indices are calculated to reflect vegetation density and growth. At the same time, the combustible material load is estimated based on remote sensing inversion models.

[0034] (3) Meteorological characteristics Meteorological features capture the weather environment of the grid on a specific date. Spatial interpolation maps the reanalysis data to each grid, thereby obtaining dynamic parameters such as temperature, humidity, wind speed, precipitation, and surface evaporation.

[0035] After completing the feature extraction and calculations described above, each grid cell is transformed into a feature vector containing dozens of specific values. The feature vectors of all grid cells are combined to form a structured dataset required for subsequent machine learning model training and prediction.

[0036] S3, input the multidimensional data features into the pre-trained XGBoost machine learning model to obtain the fire risk value of the area, and map the fire risk value of the area to the corresponding mountain power grid to obtain the fire risk value of the mountain power grid.

[0037] In a further preferred embodiment, the training process of the XGBoost machine learning model is as follows: Using the grid locations of historical thermal power plants and their corresponding multidimensional data features as positive samples, and the grid locations of unburned plants and their corresponding multidimensional data features as negative samples, the XGBoost machine learning model is trained using these positive and negative samples. This XGBoost machine learning model learns the complex relationship between features and fire by sequentially constructing multiple decision trees. Subsequent trees continuously correct the errors of previous trees, and finally, by integrating the prediction results of all trees, a high-precision risk value between 0 and 1 is output. By inputting current or predicted grid feature data into this trained model, the quantified fire risk value for each future 10m grid can be predicted.

[0038] In further optimized solutions, such as Figure 2 As shown, the mountain power grid fire prediction method also includes: Step 1: Compare the fire risk value with the preset classification threshold to obtain the fire risk level; classify the levels according to the calculated risk value of each grid, and render them with different colors on the electronic map to generate an intuitive "refined zoning map of fire risk in mountainous power grids".

[0039] Step 2: Map and issue early warnings for the corresponding mountain power grid fire risk values ​​based on the fire risk levels. To directly apply this method to power grid wildfire prevention and control, the aforementioned 10m-level grid risk data is overlaid with power grid line spatial data: line sections are divided using transmission and distribution towers as nodes, and the risk values ​​of all grid crossings within each section are combined to generate refined early warning information for specific line areas.

[0040] This application adopts: (1) Multidimensional factor coupling mechanism Wildfire risk is not determined by a single factor, but is a complex system composed of three major categories of disaster-causing factors: meteorology, geography, and vegetation. There are complex and nonlinear interactions among these factors, and assessing only a single factor is insufficient to accurately reflect the overall risk. Therefore, it is necessary to combine systems engineering theory to analyze wildfire risk assessment under the coupled conditions of meteorological, vegetation, and geographical factors.

[0041] (2) Geospatial Analysis Scale The accuracy of risk assessment is closely related to the spatial scale. Risk details that are averaged out at the kilometer level will be improved at the 10-meter level. Refining the assessment unit to a 10-meter grid is a major prerequisite for accurately locating risk sources.

[0042] (3) Ensemble learning in machine learning The XGBboost algorithm is based on the ensemble learning theory that constructs multiple prediction models (decision trees) sequentially, and allows each subsequent model to correct the residuals (errors) of the preceding models. Finally, the prediction results of these weak models are weighted and integrated into a strong prediction model with a prediction accuracy far higher than any single model.

[0043] Therefore, this invention addresses the problems of insufficient spatial accuracy, single consideration of disaster-causing factors, and insufficient risk quantification capability in existing mountain power grid fire risk prediction technologies. It provides a refined zoning and quantitative prediction method for mountain power grid risks. Through multi-source data fusion, ten-meter-level grid zoning, and XGBoost machine learning risk quantification prediction, it achieves accurate positioning, quantitative assessment, and dynamic prediction of fire risks in mountain power grid transmission and distribution channels. This significantly improves the pertinence and foresight of power grid wildfire prevention and control, and ensures the safe and stable operation of the power grid.

[0044] In summary, as follows: Figure 3 and Figure 4As shown, the overall process of this application can be divided into three stages. The first stage is data fusion and gridding, which is used to collect multi-source data, preprocess the data, and divide it into grids. The second stage is to extract features based on the preprocessed multi-source data and train the XGBoost model (quantitative risk calculation model). The third stage is dynamic prediction and application, which uses real-time or forecast data values ​​to train the XGBoost model to obtain the fire risk value of each grid, generate a refined fire risk zoning map, and finally map it to the power grid lines.

[0045] Example 2: Based on the same inventive concept, this invention also provides a mountain power grid fire prediction system, such as... Figure 5 As shown, it includes: The grid division module is used to divide the area where the power grid is located in mountainous areas into grids; The feature extraction module is used to extract multidimensional data features corresponding to each grid. The calculation module is used to input the multidimensional data features into a pre-trained XGBoost machine learning model to obtain the fire risk value of the area, and to map the fire risk value of the area to the corresponding mountain power grid to obtain the fire risk value of the mountain power grid. The network scale is no larger than 10m×10m; the multidimensional data features include terrain features, vegetation features and meteorological features.

[0046] In a further preferred embodiment, the training process of the XGBoost machine learning model in the computing module is as follows: The XGBoost machine learning model is trained using the grid where the historical thermal power plant is located and the corresponding multidimensional data features as positive samples, and the grid where the non-ignition location is located and the corresponding multidimensional data features as negative samples.

[0047] In a further preferred embodiment, the mountain power grid fire prediction system also includes a classification module, which is used for: The fire risk level is obtained by comparing the fire risk value with a preset classification threshold. Based on the fire risk level, the corresponding mountain power grid fire risk value is mapped and an early warning is issued.

[0048] In a further preferred embodiment, after obtaining the fire risk level, the risk level classification module also includes visually displaying the risk level on an electronic map using different colors.

[0049] In a further preferred embodiment, the terrain features in the feature extraction module are obtained by extracting terrain data, which includes at least one of elevation, slope, and aspect. The meteorological characteristics are obtained by extracting meteorological data, which includes at least one of surface temperature, soil moisture, cumulative precipitation, and wind direction and speed. The vegetation features are obtained by extracting vegetation data, which includes at least one of tree species, combustible load, and leaf index, wherein the combustible load is obtained by tree species inversion.

[0050] In a further preferred embodiment, the mountain power grid fire prediction method also includes a preprocessing module for resampling the meteorological data, terrain data, and vegetation data to a 10-meter resolution.

[0051] In a further preferred embodiment, the preprocessing module employs a bilinear interpolation algorithm for resampling.

[0052] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. A method for predicting power grid fires in mountainous areas, characterized in that, include: Grid division is carried out for the area where the power grid is located in the mountainous region; Extract the multidimensional data features corresponding to each grid; The multidimensional data features are input into a pre-trained XGBoost machine learning model to obtain the fire risk value of the area, and the fire risk value of the area is mapped to the corresponding mountain power grid to obtain the fire risk value of the mountain power grid. The network scale is no larger than 10m×10m; the multidimensional data features include terrain features, vegetation features and meteorological features.

2. The method for predicting power grid fires in mountainous areas according to claim 1, characterized in that, The training process of the XGBoost machine learning model is as follows: The XGBoost machine learning model is trained using the grid where the historical thermal power plant is located and the corresponding multidimensional data features as positive samples, and the grid where the non-ignition location is located and the corresponding multidimensional data features as negative samples.

3. The method for predicting power grid fires in mountainous areas according to claim 1, characterized in that, The method for predicting power grid fires in mountainous areas also includes: The fire risk level is obtained by comparing the fire risk value with a preset classification threshold. Based on the fire risk level, the corresponding mountain power grid fire risk value is mapped and an early warning is issued.

4. The method for predicting power grid fires in mountainous areas according to claim 3, characterized in that, After obtaining the fire risk level, the process also includes visually displaying the risk level on an electronic map using different colors.

5. The method for predicting power grid fires in mountainous areas according to claim 1, characterized in that, The terrain features are obtained by extracting terrain data, which includes at least one of elevation, slope, and aspect. The meteorological characteristics are obtained by extracting meteorological data, which includes at least one of surface temperature, soil moisture, cumulative precipitation, and wind direction and speed. The vegetation features are obtained by extracting vegetation data, which includes at least one of tree species, combustible load, and leaf index, wherein the combustible load is obtained by tree species inversion.

6. The method for predicting power grid fires in mountainous areas according to claim 5, characterized in that, The mountain power grid fire prediction method also includes resampling the meteorological data, terrain data, and vegetation data to a 10-meter resolution.

7. The method for predicting power grid fires in mountainous areas according to claim 6, characterized in that, Resampling is performed using a bilinear interpolation algorithm.

8. A mountain power grid fire prediction system, characterized in that, include: The grid division module is used to divide the area where the power grid is located in mountainous areas into grids; The feature extraction module is used to extract multidimensional data features corresponding to each grid. The calculation module is used to input the multidimensional data features into a pre-trained XGBoost machine learning model to obtain the fire risk value of the area, and to map the fire risk value of the area to the corresponding mountain power grid to obtain the fire risk value of the mountain power grid. The network scale is no larger than 10m×10m; the multidimensional data features include terrain features, vegetation features and meteorological features.

9. The mountain power grid fire prediction system according to claim 8, characterized in that, The training process of the XGBoost machine learning model in the computing module is as follows: The XGBoost machine learning model is trained using the grid where the historical thermal power plant is located and the corresponding multidimensional data features as positive samples, and the grid where the non-ignition location is located and the corresponding multidimensional data features as negative samples.

10. The mountain power grid fire prediction system according to claim 8, characterized in that, The mountain power grid fire prediction system also includes a classification module, which is used for: The fire risk level is obtained by comparing the fire risk value with a preset classification threshold. Based on the fire risk level, the corresponding mountain power grid fire risk value is mapped and an early warning is issued.

Images