A forest grassland fire danger intelligent early warning method, system, device and medium

Abstract

The application relates to a forest and grassland fire risk intelligent early warning method, system, device and medium. The method comprises the following steps: acquiring meteorological data and combustible moisture content and other multi-source heterogeneous data of a target area; extracting and fusing meteorological features, terrain features and vegetation features to form fire risk feature data; performing grid processing and weighted calculation on the feature data to obtain a real-time fire risk index and perform grade mapping, and combining historical data and weather forecasts to perform timing prediction to obtain a future fire risk grade; automatically identifying a high-risk area based on an evaluation result to generate an analysis report containing risk distribution and prevention measure suggestions; and matching an early warning template according to an analysis result to push precise early warning information to relevant areas and objects. The method can effectively solve the problems of insufficient data fusion, prediction lag and extensive early warning in the traditional early warning mode through multi-source data deep fusion, dynamic evaluation and intelligent analysis, and improve the active prevention capability of forest and grassland fires.

Description

Technical Field

[0001] This invention belongs to the field of forest and grassland fire early warning technology, and specifically relates to an intelligent early warning method, system, equipment and medium for forest and grassland fire risk. Background Technology

[0002] With the development of information technology in the field of forest and grassland fire prevention, various automatic weather monitoring stations and remote sensing technologies have emerged, enabling real-time monitoring of environmental parameters in key areas.

[0003] Traditional techniques typically rely on meteorological monitoring equipment distributed in key forest areas to obtain meteorological data such as temperature, humidity, wind speed, and precipitation. Preliminary assessments of forest and grassland fire risk levels are then made based on single meteorological indicators or simple empirical models. Alternatively, fire verification can be conducted solely through manual patrols and satellite remote sensing images.

[0004] However, current fire risk warning methods or traditional approaches suffer from low data integration and delayed forecasting. The lack of comprehensive consideration of multi-source, heterogeneous data, such as combustible material moisture content, complex topographic features, and vegetation type distribution, results in incomplete fire risk factor analysis and an inability to accurately reflect the actual fire risk situation. Furthermore, traditional methods are largely limited to static monitoring of the current fire risk status, lacking the ability to dynamically predict future fire risk trends. Warning information is also often limited to general level indications, failing to generate detailed risk distribution maps and specific preventative measures, thus restricting the improvement of forest and grassland fire prevention capabilities. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, system, equipment, and medium for intelligent early warning of forest and grassland fire risk, addressing the aforementioned technical issues.

[0006] Firstly, this application provides a method for intelligent early warning of forest and grassland fire risk, including: S1. Acquire multi-source heterogeneous data of forests and grasslands in the target area; among which, multi-source heterogeneous data includes meteorological data and combustible moisture content detection data; S2. Extract fire risk factor features from multi-source heterogeneous data to obtain fire risk feature data; fire risk feature data includes meteorological feature data, topographic feature data, and vegetation feature data; S3. Conduct a fire risk assessment on the fire risk characteristic data to obtain fire risk assessment data; the fire risk assessment data includes the fire risk assessment level and the future fire risk prediction level. S4. Identify high-risk fire areas from the fire risk assessment data, and conduct fire risk analysis on these high-risk areas to obtain the fire risk analysis results. The fire risk analysis results include risk area data, risk distribution data, and recommendations for risk prevention measures. S5. Generate fire risk warning results based on the fire risk analysis results; the fire risk warning results are used to instruct preset equipment to display the fire risk analysis results.

[0007] In one embodiment, fire hazard factor features are extracted from multi-source heterogeneous data to obtain fire hazard feature data, including: S11. Perform meteorological feature extraction processing on meteorological data in multi-source heterogeneous data, calculate the time change rate of temperature, relative humidity, wind speed, wind direction and precipitation, and obtain meteorological feature data. S12. Based on the digital elevation model data, perform terrain factor calculation and processing, extract slope, aspect and altitude information to obtain terrain feature data; S13. Based on remote sensing image data, extract vegetation distribution and vegetation type information, and combine combustible moisture content detection data to calculate vegetation load and litter layer thickness to obtain vegetation characteristic data. S14. Perform multi-dimensional feature fusion processing on meteorological feature data, terrain feature data and vegetation feature data to obtain fire risk feature data.

[0008] In one embodiment, fire risk assessment is performed on fire risk characteristic data to obtain fire risk assessment data, including: S21. Based on the fire hazard feature data, perform gridding processing to divide the target area into grid points of preset resolution to obtain gridded fire hazard feature data; S22. Perform weighted calculation on the gridded fire risk characteristic data to obtain the real-time fire risk index for each grid. S23. Map the real-time fire risk index to a level. Based on the preset fire risk level classification standard, map the value to the corresponding fire risk level to obtain the fire risk assessment level. S24. Based on real-time fire risk index and historical fire risk data, perform time-series trend prediction processing to calculate the change of fire risk index in a specific future time period and obtain the future fire risk prediction level.

[0009] In one embodiment, the gridded fire hazard feature data is weighted and calculated to obtain the real-time fire hazard index for each grid, including: S31. Normalize the gridded fire hazard feature data to obtain the normalized feature vector; S32. Assign feature factor weights to the normalized feature vector to obtain the weight coefficients of each feature factor. Based on the weight coefficients of each feature factor, perform a weighted summation of the feature factors to obtain the preliminary fire risk index; wherein, the expression for the preliminary fire risk index is: In the formula, This is a preliminary fire risk index. The total number of characteristic factors, For the first The weight coefficients of each feature factor, For the first Normalized values ​​of each feature factor; S33. Introduce slope correction factors and aspect correction factors to perform terrain correction on the preliminary fire risk index, obtaining the corrected fire risk index; the expression for the corrected fire risk index is: In the formula, This is the revised fire risk index. This is a preliminary fire risk index. This is the slope correction factor. This is the aspect correction factor; S34. Perform smoothing filtering on the corrected fire risk index to obtain the real-time fire risk index for each grid.

[0010] In one embodiment, time-series trend prediction processing is performed based on real-time fire risk index and historical fire risk data to calculate the change in fire risk index over a specific future time period, thereby obtaining the future fire risk prediction level, including: S41. Construct a time series based on the corrected fire risk index, extract the historical fire risk change pattern for the same period, and obtain historical trend feature data; S42. Extract short-term weather forecast elements from meteorological characteristic data to obtain weather forecast data for a future preset time period and obtain forecast characteristic data. S43. Based on historical trend characteristic data and forecast characteristic data, a prediction index for future time is calculated; the expression for the prediction index for future time is: In the formula, For the future The fire risk index is predicted at any time. For the current moment The revised fire risk index, For the future The influence value of the forecast characteristics at any given time. This is a historical value for the same period. , , For dynamic adjustment coefficients, and ; S44. Based on the preset fire risk level threshold, determine the level of the future time prediction index, determine the fire risk level for the future period, and obtain the future fire risk prediction level.

[0011] In one embodiment, high-risk fire areas are identified from fire risk assessment data, and fire risk analysis is performed on these high-risk areas to obtain fire risk analysis results, including: S51. Extract grids of fire risk levels exceeding a preset threshold from fire risk assessment data, and perform spatial aggregation based on administrative boundaries to obtain candidate high-risk areas. S52. Statistically sort the fire risk index of the candidate high-risk areas, rank the candidate high-risk areas according to the severity of fire risk, and obtain the risk area data. S53. Perform environmental overlay analysis on the risk area data to obtain risk distribution data; S54. Perform association rule matching on the risk distribution data, query the preset risk prevention knowledge base, generate risk prevention measures suggestions including suggestions for setting up isolation zones and resource allocation, integrate and generate a fire risk assessment report, and obtain fire risk analysis results.

[0012] In one embodiment, a fire risk warning result is generated based on the fire risk analysis results, including: S61. Based on the risk area data in the fire risk analysis results, determine the scope and target of the early warning information to be pushed, and obtain the target push list; S62. Based on different risk levels and push recipient permissions, match the corresponding warning template and generate warning information; S63. Based on the target push list, the early warning information is simultaneously pushed to the terminal equipment in the affected area to obtain the fire risk early warning result.

[0013] Secondly, this application also provides a forest and grassland fire risk intelligent early warning system, including: The multi-source data acquisition module is used to acquire multi-source heterogeneous data of forests and grasslands in the target area; among which, the multi-source heterogeneous data includes meteorological data and combustible moisture content detection data; The fire risk feature extraction module is used to extract fire risk factor features from multi-source heterogeneous data to obtain fire risk feature data; the fire risk feature data includes meteorological feature data, terrain feature data and vegetation feature data; The fire risk level assessment module is used to assess the fire risk characteristics and obtain fire risk assessment data; the fire risk assessment data includes the fire risk assessment level and the future fire risk prediction level. The fire risk analysis module is used to identify high-risk fire areas from fire risk assessment data, and to perform fire risk analysis on these high-risk areas to obtain fire risk analysis results. The fire risk analysis results include risk area data, risk distribution data, and risk prevention measures recommendations. The fire hazard warning module is used to generate fire hazard warning results based on the fire hazard analysis results; the fire hazard warning results are used to instruct preset equipment to display the fire hazard analysis results.

[0014] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0015] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0016] The aforementioned intelligent early warning method, system, equipment, and medium for forest and grassland fire risk acquire and integrate multi-source heterogeneous data such as meteorological data and combustible material moisture content, and extract and fuse multi-dimensional fire risk features to achieve a comprehensive and dynamic characterization of factors influencing forest and grassland fire risk. Based on gridded processing and weighted calculation, combined with terrain correction and trend prediction, it can generate refined real-time fire risk level assessments and future fire risk level predictions, improving the accuracy and foresight of fire risk assessments. This method can also automatically identify high-risk areas, perform spatial aggregation and sorting analysis, and generate assessment reports containing specific risk prevention measures recommendations, ultimately achieving accurate and rapid delivery of early warning information. It effectively overcomes the shortcomings of traditional methods, such as insufficient data fusion, static and delayed judgment, and vague early warning information, forming a closed loop from data perception and intelligent analysis to accurate early warning, comprehensively improving the prevention capabilities and emergency response levels of forest and grassland fires. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating an intelligent early warning method for forest and grassland fire risk in one embodiment. Figure 2 This is a schematic diagram of the structure of an intelligent early warning system for forest and grassland fire risk in one embodiment; Figure 3 This is a schematic diagram of the structure of a computer device in one embodiment. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0020] refer to Figure 1The application presents a flowchart illustrating a method for intelligent early warning of forest and grassland fire risk, which includes the following steps: S1. Obtain multi-source heterogeneous data of forests and grasslands in the target area.

[0021] Among them, multi-source heterogeneous data includes meteorological data and combustible material moisture content detection data.

[0022] Specifically, when acquiring multi-source heterogeneous data on forests and grasslands in the target area, it is necessary to rely on comprehensive fire risk monitoring stations deployed in high-risk and high-risk counties for forest and grassland fire prevention and control to construct a comprehensive, all-weather, and full-process data acquisition network to ensure the integrity and timeliness of data coverage. Meteorological data within the multi-source heterogeneous data is acquired through highly integrated multi-element meteorological sensors. These sensors must meet industrial-grade IP67 or higher protection standards, possess rust-proof and lightning-proof capabilities, and be able to operate reliably in high and low temperature environments ranging from -40℃ to 70℃, as well as in humid and hot environments and high-radiation environments. In field application scenarios, their mean time between failures (MTBF) should be no less than 10,000 hours, and they should be able to accurately measure key meteorological parameters such as air temperature, air humidity, wind speed, wind direction, rainfall, atmospheric pressure, and illuminance. The measurement accuracy of each parameter must meet specific standards: air temperature measurement range is -40℃ to 85℃, resolution not exceeding 0.1℃, and measurement error controlled within ±0.3℃; air humidity measurement range covers 0 to 100%RH, resolution not exceeding 0.1%RH, and error not exceeding ±3%RH; wind speed measurement range is 0 to 60m / s, resolution not exceeding 0.01m / s, and error percentage not exceeding 10%; wind direction measurement range is 0 to 360°, resolution not exceeding 0.1°, and error percentage not exceeding 10%; rainfall measurement range is 0 to 200mm / h, resolution not exceeding 0.2mm, and error percentage not exceeding 10%; atmospheric pressure measurement range is 300 to 1100hpa, resolution not exceeding 0.1hpa, and error not exceeding ±0.3hPa; illuminance measurement range is 0 to 200KLUX, resolution not exceeding 10LUX, and error percentage not exceeding 3%.

[0023] Specifically, the combustible material moisture content detection data is collected by a combustible material moisture content detection sensor. This sensor supports simultaneous measurement of the surface, inner, and bottom layers of surface litter, which includes diverse types such as coniferous, broadleaf, and mixed species, with a thickness ranging from 2 cm to 10 cm. The sensor employs a multi-sensor fusion detection mode, with a moisture content measurement range of 0 to 100%. The accuracy error is no more than ±5% when the moisture content is ≤35%, and no more than ±10% when the moisture content is between 35% and 100%, with a resolution not exceeding 0.1%. Simultaneously, the sensor can collect data on surface phenology, soil moisture content, and surface temperature and humidity. The soil moisture content measurement range is 0 to 50% Vol, with a resolution not exceeding 0.1% and an accuracy error not exceeding ±5%. The surface temperature measurement range is -40℃ to 85℃, with a resolution not exceeding 0.01℃ and an error not exceeding ±0.3℃. The surface humidity measurement range is 0 to 100% RH, with a resolution not exceeding 0.04% RH and an error not exceeding ±2% RH.

[0024] The data collection and reporting frequency supports multiple options such as 1 minute, 30 minutes, and 1 hour, enabling minute-level adjustable automatic collection and reporting. The communication unit uses RS485 local communication or GPRS / 3G / 4G / 5G public network transmission. Sensing data reporting must employ an encryption algorithm of at least 128 bits to ensure transmission security from three dimensions: physical environment, network equipment, and the data itself, preventing data leakage and malicious tampering. Collected data must be connected to the provincial emergency management integrated application platform via standard protocols such as TCP and UDP to achieve data aggregation and sharing with the National Forest and Grassland Fire Prevention and Control Information Sharing Platform. Interfaces for forestry, meteorology, and other industries are reserved to meet cross-departmental data interaction needs. The initial data storage resource configuration is no less than 300GB, with an annual increase of 10GB of sensing data per station and 30TB of new data output per station. The storage period is no less than 6 months to ensure data traceability and integrity for subsequent analysis and use.

[0025] S2. Extract fire risk factor features from multi-source heterogeneous data to obtain fire risk feature data.

[0026] The fire risk characteristic data includes meteorological characteristic data, topographic characteristic data, and vegetation characteristic data.

[0027] Specifically, before extracting fire hazard factor features from the acquired multi-source heterogeneous data, a data preprocessing process must be performed to ensure the accuracy and validity of the feature data. In the preprocessing stage, the Laida criterion is used to remove outlier data, i.e., filtering data that exceeds the sensor measurement range or deviates from the mean by three times the standard deviation. Then, missing data is filled in using linear interpolation. Finally, the Z-Score standardization method is used to convert the original data of different dimensions to a uniform scale. The Z-Score standardized data has a mean of 0 and a variance of 1, which effectively avoids interference from differences in data magnitude on the feature extraction results.

[0028] The extraction of meteorological characteristic data is based on preprocessed meteorological sensor data, and the sliding window method is used to calculate time-series statistical features. Specifically, the maximum, minimum, average, and rate of change of air temperature, minimum air humidity, and duration of continuous exposure to 30% RH are extracted for 1 hour, 6 hours, 12 hours, and 24 hours; the average, instantaneous maximum, and wind direction concentration of air speed are extracted. The wind direction concentration is calculated using a vector synthesis method to determine the dispersion of wind direction distribution, thereby reflecting the impact of wind direction stability on fire spread. Simultaneously, parameters such as the cumulative rainfall over time and the number of consecutive days without rainfall, the cumulative daily duration of sunlight and its noon peak, and the slope of atmospheric pressure trends are extracted to form core meteorological characteristic data that comprehensively reflects the impact of meteorological conditions in the target area on fire risk.

[0029] The extraction of terrain feature data requires the integration of BeiDou satellite positioning data with the EGIS (Environmental Geographic Information System) base map of the Ministry of Emergency Management or local high-resolution base map services, and refined extraction is achieved through GIS spatial analysis tools. First, the target area is divided into grids with a 30-meter resolution, and the elevation data of each grid point is obtained. The Deloni triangulation method is used to calculate the slope and aspect corresponding to each grid point, with slope values ​​ranging from 0 to 90° and aspect values ​​ranging from 0 to 360°. Second, a terrain morphology recognition algorithm is used to screen out special dangerous terrain areas such as valleys between two mountains and narrow ridges. A binary variable is used to indicate whether each grid point is in a special dangerous terrain area: 1 indicates it is, and 0 indicates it is not. Simultaneously, administrative division data is combined to associate the district / county and township-level information to which each grid point belongs, providing a basic geographical basis for subsequent zonal fire risk assessment.

[0030] Vegetation characteristic data extraction is based on the fusion analysis of multispectral phenological sensor data and combustible material moisture content detection data. The normalized vegetation index is calculated using visible / near-infrared dual-band data, using the following formula: .in, It represents the reflectivity of the near-infrared band, which typically ranges from 750 nm to 900 nm and can effectively reflect the growth vitality of vegetation. Reflectance in the red light band typically ranges from 620nm to 750nm. Vegetation exhibits low reflectance in this band, which can be used to distinguish vegetated from non-vegetated areas. Phenological analysis results are combined to identify vegetation greening, yellowing, and snow cover, determining the vegetation growth stage and flammability baseline. Based on the average moisture content of the surface, inner, and bottom layers of litter and the differences in moisture content between layers, flammability characteristics of litter are extracted. Simultaneously, combined with a vegetation type distribution map, each grid point is labeled with a classification code, with coniferous forests, broad-leaved forests, and mixed forests each corresponding to different codes, forming a complete vegetation feature dataset that provides comprehensive vegetation-related data for fire risk assessment.

[0031] S3. Conduct fire risk assessment on the fire risk characteristic data to obtain fire risk assessment data.

[0032] The fire risk assessment data includes the fire risk assessment level and the future fire risk prediction level.

[0033] Specifically, the fire risk assessment process employs a combination of the Analytic Hierarchy Process (AHP) and fuzzy comprehensive evaluation, while incorporating a time series forecasting model to achieve accurate calculations of both the fire risk assessment level and the future fire risk prediction level. First, the weights of each fire risk characteristic data are determined using the AHP, constructing a three-level evaluation system: the target layer, the criterion layer, and the indicator layer. The target layer assesses the fire risk level; the criterion layer includes meteorological characteristics, topographic features, and vegetation characteristics; and the indicator layer comprises the specific characteristic parameters under each criterion layer. Ten to fifteen experts in forest and grassland fire prevention and control are invited to conduct pairwise comparisons of the importance of each level of indicator, constructing a judgment matrix. The rationality of the judgment matrix is ​​verified through a consistency test; the consistency ratio (CR) must be less than 0.1. If this requirement is not met, the expert judgments must be readjusted until the test is passed. Finally, the weight values ​​of each indicator are calculated. Among the meteorological characteristics, temperature change rate and duration of low humidity; among the topographic features, slope and special hazardous terrain markers; and among the vegetation features, litter moisture content and vegetation type coding, have relatively high weights, totaling no less than 60%, ensuring the dominant role of core fire risk factors in the assessment results.

[0034] Subsequently, the fuzzy comprehensive evaluation method was used to calculate the fire risk assessment level. Based on historical fire case data and expert experience, five fuzzy evaluation levels were set for each indicator: extremely low, low, medium, high, and extremely high, with a corresponding membership function assigned to each level. The membership function quantifies the degree to which each indicator belongs to different fire risk levels. The standardized fire risk feature data was substituted into the membership function to obtain the fuzzy evaluation matrix for each indicator. Then, fuzzy matrix multiplication was performed with the indicator weight vector to obtain the comprehensive fuzzy evaluation result. The fire risk assessment level for each 30-meter grid point was determined according to the principle of maximum membership degree, that is, the level with the highest membership degree in the comprehensive fuzzy evaluation result was selected as the final assessment level for that grid point. Finally, the grid point levels were summarized according to administrative divisions to form regional fire risk assessment levels, and the system supports descending ranking and display of the assessment levels for each region.

[0035] The calculation of future fire risk prediction levels employs a Long Short-Term Memory (LSTM) neural network model. This model possesses the ability to capture long-term dependencies in time-series data and is suitable for predictive analysis of fire risk characteristic time-series data. The model input consists of hourly fire risk characteristic time-series data for the target area over the past 30 days, including dynamic changes in meteorological and vegetation characteristics, as well as static terrain characteristic data. The static terrain characteristic data is adapted to the time-series input format through repeated padding. The model output provides hourly fire risk predictions for the next 24 hours and daily fire risk predictions for the next 7 days, with prediction levels categorized into five levels: extremely low, low, medium, high, and extremely high.

[0036] During model training, three years of historical fire risk monitoring data and corresponding fire occurrence data were used as the dataset, divided into training, validation, and test sets in a 7:1.5:1.5 ratio. The cross-entropy loss function was used as the optimization objective, and its expression is: .in, Indicates the number of samples. Indicates the number of fire hazard categories; Indicates the first The sample belongs to the first The true label of the category level is represented in one-hot encoding form, that is, it is 1 if it belongs to the category, and 0 otherwise. The model predicts the first... The sample belongs to the first The probability of class-level fire risk assessment is determined. Model parameters are adjusted using the Adam optimizer, with the initial learning rate set to 0.001. This learning rate adaptively decreases with each training epoch to ensure the model converges to its optimal state, achieving a final prediction accuracy of at least 85%. During the evaluation process, data product generation time must not exceed 5 seconds. If processing time is extended due to special circumstances such as a surge in data volume, progress must be displayed in real-time using a progress bar or percentage. The system also supports kilometer-level fine-grained grid early warning, with a grid resolution strictly maintained at at least 30 meters. It adapts to the hierarchical access requirements of web applications, based on provincial, municipal, and county-level access control rules. This ensures that users at different levels can only view fire risk assessment data and prediction results within their corresponding administrative regions, guaranteeing data access security and relevance.

[0037] S4. Identify high-risk fire areas from the fire risk assessment data, and conduct fire risk analysis on these high-risk areas to obtain the fire risk analysis results.

[0038] The fire risk analysis results include risk area data, risk distribution data, and recommendations for risk prevention measures.

[0039] Specifically, the identification of high-risk fire areas is based on fire risk assessment data, with a clear and unified identification standard: grid areas with a fire risk assessment level of "high" or "extremely high," and a fire risk prediction level remaining "high" or above for the next 4 hours, are defined as high-risk fire areas. The DBSCAN spatial clustering algorithm aggregates adjacent high-risk grid points into continuous risk areas. This algorithm does not require pre-setting the number of clusters and can automatically identify clusters of any shape. During the clustering process, a neighborhood radius of 30 meters is set (consistent with the grid resolution), and a minimum number of grid points is set to 9 to ensure the continuity and rationality of the aggregated risk areas. The latitude and longitude boundaries, area, and administrative divisions involved in each high-risk area are determined, with administrative divisions accurate to the village level. Key protected targets within the area are also marked, including nature reserves, forest parks, ancient buildings, and residential areas, forming complete risk area data.

[0040] The generation of risk distribution data requires the integration of high-resolution base map services and risk area information. A color gradient coding rule is used to display risk levels, with different levels corresponding to different colors: extremely low risk is dark green, low risk is light green, medium risk is yellow, high risk is red, and extremely high risk is dark red. A risk area layer is overlaid on the EGIS base map, marking the level, area, and core risk points of each risk area. Core risk points include areas with extremely low litter moisture content and steep coniferous forest areas. The generated refined risk distribution map supports layer scaling, panning, and transparency adjustment. It can be overlaid with auxiliary layers such as vegetation type, terrain, and emergency resource distribution. The grid resolution is maintained at no less than 30 meters to ensure the visualization accuracy of risk distribution and facilitate a direct understanding of the regional fire risk distribution.

[0041] Fire risk analysis also requires the integration of 3D GIS maps for refined fire situation assessment. This involves consolidating multi-dimensional information on high-risk areas and their surroundings, including real-time meteorological information such as hourly wind speed, wind direction, temperature, and humidity; geographical information such as vegetation type distribution, altitude, slope aspect, and location of special hazardous terrain; emergency resource data such as fire station locations, fire water source distribution, number of emergency teams, and equipment configuration; and population size and distribution data for the district. The Rothermel fire spread model is used to calculate the possible spread range, spread speed, and spread direction of the fire. The expression for the fire spread speed is: .in, This indicates the rate of fire spread, expressed in m / min. The ignition coefficient is related to the moisture content of the combustible and the type of vegetation. The lower the moisture content and the higher the proportion of flammable vegetation, the greater the ignition coefficient. This represents the fuel bed width coefficient, reflecting the influence of fuel bed width on the spread rate; This represents the fuel bed load factor, which is the weight of combustible material per unit area, expressed in kg / m². This represents the slope coefficient; the steeper the slope, the faster the spread. It is calculated from the slope value. This represents the wind speed coefficient, which is related to the wind speed, the angle between the wind direction and the slope. It represents the fuel continuity coefficient, reflecting the degree of continuity in the distribution of combustibles.

[0042] Based on different meteorological scenarios, the simulation shows changes in fire spread trends, including common scenarios such as increased wind speed and changes in wind direction, analyzing the impact of scenario changes on the range and speed of fire spread. It also supports setting firebreaks with different parameters, with widths of 5 meters, 10 meters, and 15 meters, and their locations can be flexibly set according to terrain and risk point distribution. The simulation demonstrates the firebreak's effectiveness in blocking fire spread and estimates the fire control area and response time under different treatment plans. Risk prevention measures should be tailored to the specific characteristics of the risk area: For high-risk areas in coniferous forests with slopes ≥30°, it is recommended to increase the number of professional patrol teams, increase the patrol frequency to once every 2 hours, clear fallen leaves to a thickness of ≤5cm, and set up temporary firebreaks ≥10 meters wide at the edge of the area; For high-risk areas near residential areas, it is recommended to remind residents through village broadcasts and door-to-door notifications by grid workers to clear flammable materials within 5 meters of their houses, prepare basic fire-fighting equipment such as fire extinguishers and buckets, and strictly prohibit open fires; For high-risk areas involving key protected targets, it is recommended to deploy emergency response teams in advance, equip them with professional equipment such as high-pressure water pumps and fire hoses, and develop specific emergency response plans to ensure that risk prevention measures are operable and targeted, forming a complete set of risk prevention measures recommendations.

[0043] S5. Generate fire risk warning results based on the fire risk analysis results.

[0044] Among them, the fire risk warning result is used to instruct the preset equipment to display the fire risk analysis result.

[0045] Specifically, the generation of fire risk warning results requires the integration of risk area data, risk distribution data, and risk prevention measure recommendations to form a multi-format warning data package compatible with multiple preset devices, ensuring efficient transmission and visualization of warning information. The warning results comprise three core modules: visualized data, multi-channel alarm information, and standardized fire risk assessment reports, meeting the warning needs of different scenarios.

[0046] The visualized data display needs to be compatible with the display requirements of preset devices such as web terminals, emergency command center large screens, and mobile apps, and supports the visualized distribution display of comprehensive fire risk factor monitoring stations. By clicking the monitoring station icon, users can view the station's real-time monitoring data, as well as historical data for fixed time periods such as 30 minutes, 1 hour, 2 hours, 6 hours, 12 hours, and 24 hours. Users can also customize time periods to query historical data. Query results can be exported and downloaded in Excel and CSV formats for easy data analysis and archiving. The risk distribution map supports high-definition display on various preset devices. Users can filter and view the risk distribution of corresponding areas by administrative division, covering the provincial, municipal, county, and township levels. By clicking on a risk area, users can view detailed information, including risk level, area, affected villages and towns, key protected targets, and spread prediction results. A timeline control is also supported, dynamically displaying the hourly risk distribution trend for the next 24 hours, intuitively presenting the evolution of fire risk patterns.

[0047] Multi-channel alerts are targeted to high-risk areas, with pre-set devices including mobile phones and office computers of emergency management personnel, rural broadcasting equipment, scenic area voice broadcasting systems, and large screens in the emergency command center. Alerts are sent to relevant personnel via SMS gateways, including the name, level, affected area, estimated duration, and core prevention requirements of the high-risk area, ensuring personnel quickly grasp key information. Alerts are converted into audio broadcasts using speech synthesis technology and played repeatedly through rural broadcasting and scenic area broadcasting systems, covering the public and staff within the area and improving alert coverage. Pop-up alerts and video alerts are pushed to web and mobile apps. Video content includes a visual display of the risk area and explanations of prevention measures. Pop-up alerts use prominent colors to ensure relevant personnel receive alert information immediately, avoiding delays in response.

[0048] The standardized fire risk assessment report is generated in real-time in PDF format, with a standardized structure and detailed content. The report includes four parts: a risk zoning list, a risk distribution map, recommendations for risk prevention measures, and recommendations for emergency resource allocation. The risk zoning list is sorted by risk level, listing basic information, core risk factors, and the basis for risk level determination for each risk area. The risk distribution map includes a high-resolution, printable risk distribution map and a spread trend prediction map, with clear annotations and sufficient accuracy, suitable for direct use in emergency command and decision-making. Recommendations for risk prevention measures are developed by region and scenario, clearly defining responsibilities and execution deadlines to ensure the feasibility of the measures. The emergency resource allocation recommendations, based on emergency resource distribution data, recommend the nearest fire stations, emergency teams, and equipment configurations, supporting efficient resource allocation. The report supports online viewing and download on various preset devices, adapting to the needs of different users.

[0049] The data transmission of fire risk warning results adopts the standard TCP / UDP protocol and connects to the provincial emergency management integrated application platform through an encrypted channel. This enables real-time data synchronization with the national forest and grassland fire prevention and control information sharing platform. Interfaces for forestry, meteorology, and other industries are reserved to support cross-departmental data exchange. The display of pre-set devices must follow a three-tiered access control system: provincial users can view all warning data across the province, including risk analysis results and warning information for each city and county level, and have data export and statistical analysis permissions; city-level users can view warning data for their city and subordinate county-level areas, supporting regional data aggregation and comparative analysis; county-level users can view warning data for their county and township-level areas, focusing on displaying detailed information and preventative measures for high-risk areas within their jurisdiction. This ensures that users at all levels can obtain accurate warning services according to their permissions, supporting efficient prevention and emergency response to forest and grassland fire risks.

[0050] The aforementioned intelligent early warning method for forest and grassland fire risk integrates multi-source heterogeneous data such as meteorological data, combustible material moisture content, topographic data, and vegetation data to achieve comprehensive collection and deep fusion of forest and grassland fire risk factors, overcoming the limitations of traditional methods that rely on single data sources and insufficient fusion. Based on multi-dimensional features, it performs gridded dynamic assessment and future trend prediction, improving the precision and timeliness of fire risk level determination. By intelligently identifying high-risk areas and generating analysis reports containing risk distribution maps and specific preventative measures, it achieves a leap from macro-level alerts to precise spatial guidance in early warning information. Through a hierarchical and categorized early warning information push mechanism, it effectively improves the targeting and efficiency of early warning responses, thereby enhancing the early prevention and precise control capabilities of forest and grassland fires.

[0051] In an optional embodiment, fire hazard factor features are extracted from multi-source heterogeneous data to obtain fire hazard feature data, including the following steps: S11. Perform meteorological feature extraction processing on meteorological data in multi-source heterogeneous data, calculate the time change rate of temperature, relative humidity, wind speed, wind direction and precipitation, and obtain meteorological feature data.

[0052] Optionally, meteorological feature extraction should be based on preprocessed hourly meteorological monitoring data, using a combination of fixed and sliding time windows to calculate time-series features, ensuring accurate reflection of the dynamic changes in meteorological parameters. The calculation of the time-series temperature change rate is based on a set time interval, using linear regression to fit the temperature time-series curve within the time period. The slope represents the temperature change rate for the corresponding time period. Simultaneously, the maximum, minimum, and average temperatures for each time period, as well as the cumulative temperature rise, are calculated. The cumulative temperature rise is the difference between the highest temperature and the initial temperature within the time period, used to characterize the risk of a sudden short-term temperature increase. The time-series relative humidity change rate is calculated at the same time intervals, focusing on the rate of humidity decrease. When the humidity change rate reaches a set threshold, it is marked as a high-risk feature point. The minimum humidity value, the duration of sustained low humidity, and the humidity fluctuation variance are also statistically analyzed for each time period. A smaller fluctuation variance indicates more stable humidity and a longer duration of flammable environment.

[0053] The wind speed temporal change rate is calculated using data filtered by a moving average to avoid interference from sudden changes in instantaneous wind speed. By setting a sliding window size to filter the data, the wind speed change rate is calculated for different time periods. The mean wind speed, instantaneous maximum wind speed, and duration of sustained wind speed exceeding the critical value are extracted for each time period. A longer duration of wind speed exceeding the critical value indicates a higher risk of fire spread. The wind direction temporal change rate is calculated using vector angle changes. Wind direction is converted to an angle value, and the change in wind direction between adjacent time periods is calculated using the angle difference formula. This change is then divided by the time interval to obtain the change rate. Simultaneously, the wind direction concentration is calculated using a vector synthesis method. A smaller wind direction change rate and higher concentration make it easier to predict the direction of fire spread. The precipitation temporal change rate is calculated for different time periods, using the ratio of the cumulative precipitation difference to the time interval. Rainfall intensity levels are differentiated, and the number of consecutive days without rainfall and the precipitation decay trend during the post-rain recovery period are statistically analyzed. The precipitation change rate during the post-rain recovery period directly affects the rate of recovery of combustible material moisture content and is a key factor in fire risk assessment. After the time-varying rate of change of all meteorological parameters is calculated, it is integrated with the statistical characteristics of each time period to form meteorological characteristic data with unified dimensions and complete time series. The data format adopts a standardized matrix form to facilitate subsequent fusion processing.

[0054] S12. Based on the digital elevation model data, perform terrain factor calculation and processing to extract slope, aspect and altitude information to obtain terrain feature data.

[0055] Optionally, terrain feature extraction relies on digital elevation model (DEM) data at a set resolution. The data source can be a fusion of satellite remote sensing mapping data and ground-based measured data. The DEM data is preprocessed first, using a depression-filling algorithm to eliminate abnormal depressions and employing smoothing filtering to remove noise interference, ensuring the continuity and accuracy of the terrain data. Slope extraction uses the third-order difference method to calculate the slope value of each grid point. The ratio of the elevation difference to the horizontal distance between adjacent grid points in the DEM data is converted into a slope angle, and feature intervals are divided according to slope grade. Different slope grades correspond to different fire spread rate weights; the steeper the slope, the faster the fire spreads.

[0056] Slope aspect extraction is determined by calculating the angle between the slope normal at each grid point and true north, with a value ranging from 0 to 360°, and divided into multiple azimuth intervals. South-facing slopes, due to longer sunshine hours, higher surface temperatures, and lower moisture content of combustibles, are more flammable and require key feature identification; north-facing slopes, with shorter sunshine hours and higher humidity, are relatively less flammable. Elevation information is directly extracted from the preprocessed digital elevation model data, with each grid point corresponding to a unique elevation value. The elevation range, average elevation, and elevation gradient within the region are also calculated. The elevation gradient reflects the degree of topographic relief; greater relief hinders fire fighting and requires separate labeling in the feature data. After topographic factor extraction, slope, aspect, elevation, and derived features are associated with grid point coordinates to form topographic feature data corresponding to the target area grid. The topographic features of each grid point are stored in vector form to ensure spatial matching with subsequent meteorological and vegetation features.

[0057] S13. Based on remote sensing image data, extract vegetation distribution and vegetation type information, and combine combustible moisture content detection data to calculate vegetation load and litter layer thickness to obtain vegetation characteristic data.

[0058] Optionally, vegetation information extraction utilizes multispectral remote sensing image data at a set resolution, acquiring images at fixed intervals to ensure timely capture of changes in vegetation growth status. Vegetation distribution information extraction is achieved by calculating the normalized vegetation index (NVI). Based on the NVI values, vegetation cover types are classified, thereby generating a vegetation cover distribution map to clearly define the spatial distribution range of vegetation.

[0059] Vegetation type information extraction employed a supervised classification algorithm, combined with a field vegetation survey sample database, to classify vegetation into multiple types. After classification, accuracy verification was performed, and a confusion matrix was used to correct classification errors, ensuring accurate vegetation type identification. Vegetation carrying capacity calculation utilized a normalized vegetation index and biomass regression model. A regression equation was constructed based on historical measured vegetation biomass data, and the mean and maximum vegetation carrying capacity were statistically analyzed for each grid point. Litter layer thickness calculation combined combustible material moisture content detection data with vegetation type, achieved through a three-way correlation model. Different vegetation types exhibit different litter accumulation rates. The theoretical litter thickness was initially estimated using vegetation growth years retrieved from remote sensing imagery, and then corrected using measured moisture content data. Lower moisture content indicates drier and more compressed litter, resulting in a smaller thickness correction coefficient. The actual litter layer thickness at each grid point was obtained. Integrating vegetation distribution, type, carrying capacity, and litter thickness information formed vegetation characteristic data.

[0060] S14. Perform multi-dimensional feature fusion processing on meteorological feature data, terrain feature data and vegetation feature data to obtain fire risk feature data.

[0061] Optionally, feature fusion employs a feature-level fusion strategy. First, spatial alignment of the three types of feature data is performed. Using the set grid coordinates as a reference, meteorological, topographic, and vegetation features are mapped to the same spatial grid, ensuring that the three types of features at each grid point are matched one-to-one, eliminating the impact of spatial location deviations on the fusion result. Subsequently, the features of each dimension are normalized, using the Min-Max normalization method to map all feature values ​​to a specified interval, avoiding imbalances in fusion weights due to differences in feature dimensions.

[0062] The fusion process employs a combination of weighted fusion and principal component analysis. First, the weight proportions of meteorological, topographical, and vegetation features are determined using the analytic hierarchy process (AHP). Then, weights are assigned to sub-features within each feature category. The weight proportion of core sub-features is no less than a predetermined proportion of the total weight of their respective categories. After weighted summation, an initial fused feature matrix is ​​obtained. Principal component analysis is then used to reduce feature dimensionality, remove redundant information, and retain principal components with a predetermined cumulative contribution rate as the final fire risk feature data. This reduces the computational load of subsequent models while preserving core fire risk information. After fusion, the fire risk feature data is validated by removing grid data with abnormal feature values ​​and filling in missing values ​​using spatial interpolation to ensure the integrity and reliability of the feature data. The final output fire risk feature data is stored in the form of grid vectors, with each grid point corresponding to a set of fused feature values, providing comprehensive and accurate input data for subsequent fire risk assessment.

[0063] In an optional embodiment, fire risk assessment is performed on the fire risk characteristic data to obtain fire risk assessment data, including the following steps: S21. Based on the fire hazard feature data, perform gridding processing to divide the target area into grid points of preset resolution to obtain gridded fire hazard feature data.

[0064] Optionally, the gridding process uses the spatial coordinates of the fire hazard feature data as a reference, employing a square grid to divide the target area. The preset resolution is consistent with the grid resolution extracted from the terrain and vegetation features mentioned earlier, ensuring spatial matching of the data. During the division process, the grid range is first determined based on the latitude and longitude boundaries of the target area. The latitude and longitude coordinates are then transformed into Cartesian coordinates using the Gaussian projection coordinate system, and the total number of rows and columns of the grid is calculated. Boundary areas with less than one grid point are padded with complete grid points to avoid data omission.

[0065] When assigning grid values ​​to fire risk feature data, spatial interpolation is used to handle the matching relationship between feature data and grids. For grids with multiple feature data points, the average feature value is taken as the feature value of that grid. For grids without feature data points, Kriging interpolation is used to supplement them, and an interpolation model is constructed based on the feature data of neighboring grids to ensure interpolation accuracy. Each grid is assigned a unique identifier number to facilitate subsequent risk tracing and location. Topographic static feature data, vegetation type data, etc., are associated with grids one by one to form gridded fire risk feature data where each grid contains multi-dimensional features of meteorology, topography, and vegetation. The data is stored in matrix form, with rows corresponding to grid row numbers and columns corresponding to feature types, providing structured input for subsequent weighted calculations.

[0066] S22. Perform weighted calculation on the gridded fire hazard characteristic data to obtain the real-time fire hazard index for each grid.

[0067] Optionally, before weighted calculation, the weight coefficients of each characteristic indicator need to be determined. The weights are determined using the analytic hierarchy process (AHP), following the three-level evaluation system of target layer, criterion layer, and indicator layer. Experts in forest and grassland fire prevention and control are invited to compare the importance of each indicator pairwise, construct a judgment matrix, and pass a consistency test to finally determine the weights of the criterion layer. The weights of core characteristics of the indicator layer, such as temperature change rate, slope, and vegetation load, should not be less than a preset threshold in the total weight of their respective criterion layers. The weighted calculation uses a linear weighted summation formula. During the calculation process, the standardized characteristic values ​​of each grid point are multiplied by their corresponding weights, and then summed to obtain the real-time fire risk index for that grid point. At the same time, an outlier filtering mechanism is set up. If any characteristic value of a grid point exceeds the normal range (±3 standard deviations), the grid point is removed and the index is supplemented by the mean of neighboring grid points to ensure the accuracy of the real-time fire risk index.

[0068] S23. Map the real-time fire risk index to a level. Based on the preset fire risk level classification standard, map the value to the corresponding fire risk level to obtain the fire risk assessment level.

[0069] Optionally, the fire risk level is pre-defined into five levels: extremely low risk, low risk, medium risk, high risk, and extremely high risk, clearly defined according to real-time fire risk index ranges. This classification standard is calibrated based on historical fire case data, and the threshold values ​​of each range are adjusted by statistically analyzing the probability of fire occurrence within different index ranges to ensure that the probability of fire occurrence corresponding to each level meets actual prevention and control needs.

[0070] During the fire risk level mapping process, an interval matching method is used to map the real-time fire risk index of each grid point to a specific fire risk level. Simultaneously, a fuzzy membership function is used to optimize the boundary interval mapping results. For grid points where the index lies at the interval boundary, the final level is determined through membership degree calculation, and the level with the higher membership degree is taken as the final assessment level to avoid subjectivity in boundary value mapping. After mapping, the grid point levels are summarized and statistically analyzed according to administrative divisions, and the area proportion of different fire risk levels in each region is calculated to form the regional fire risk assessment level. The distribution locations of core high-risk grid points in each region are also marked, providing a basis for subsequent high-risk area identification.

[0071] S24. Based on real-time fire risk index and historical fire risk data, perform time-series trend prediction processing to calculate the change of fire risk index in a specific future time period and obtain the future fire risk prediction level.

[0072] Optionally, the time series trend prediction adopts a long short-term memory neural network (LSTM) model. The model input data includes two parts: one is the hourly gridded real-time fire risk index time series data of the target area over a period of time, and the other is the historical fire risk characteristic data of the corresponding time period, namely the dynamic change data of meteorological, topographic and vegetation characteristics. The static topographic feature data is adapted to the time series input format by repeated filling to ensure the uniformity of the input data dimensions.

[0073] For example, a specific future time period can be set as the next 24 hours (hourly prediction) and the next 7 days (daily prediction). The model training uses three years of historical fire risk monitoring data and fire occurrence data as the dataset, which is divided into training, validation, and test sets proportionally. The cross-entropy loss function is used as the optimization objective, and the model parameters are adjusted using the Adam optimizer until the model converges. During the prediction process, the predicted fire risk index value for each grid point within the specific future time period is first calculated. Then, based on the set level classification criteria, the predicted index is mapped to the corresponding fire risk level to obtain the future fire risk prediction level.

[0074] Meanwhile, an error correction mechanism is employed to optimize the forecast results, and the forecast index is adjusted in conjunction with real-time meteorological forecast data to ensure that the forecast results closely match the actual meteorological change trends. The final output of the future fire risk forecast level must be labeled with the forecast credibility. The credibility is calculated based on the model forecast error; the smaller the error, the higher the credibility. High-credibility forecast results are given priority as the basis for prevention and control, providing accurate time-series guidance for fire risk prevention work.

[0075] In an optional embodiment, the gridded fire hazard feature data is weighted and calculated to obtain the real-time fire hazard index for each grid, including the following steps: S31. Normalize the gridded fire hazard feature data to obtain the normalized feature vector.

[0076] Optionally, normalization is performed separately for each feature factor in the gridded fire risk feature data. The core purpose is to eliminate the weight imbalance caused by the difference in the dimensions of different feature factors, so that all feature factors are on the same numerical scale for subsequent weighted calculations. The Min-Max normalization method is used to achieve normalization. This method can preserve the relative distribution relationship of feature factors, adapt to the statistical characteristics of fire risk feature data, and ensure that it adapts to the feature fluctuations caused by vegetation growth cycles and seasonal weather changes.

[0077] During the normalization process, an outlier detection mechanism is implemented simultaneously. For raw data that exceeds the normal range of feature factors, the mean of neighboring grid points is used to replace the original data before standardization calculation, thus avoiding distortion of the normalization result due to outliers. After all feature factors of each grid point are normalized, they are combined according to a preset feature order to form the normalized feature vector of that grid point. The vector dimension is consistent with the total number of feature factors, and the vector format is stored as a row vector for easy subsequent matrix operations and weight allocation.

[0078] S32. Assign feature factor weights to the normalized feature vector to obtain the weight coefficients of each feature factor. Based on the weight coefficients of each feature factor, perform a weighted summation of the feature factors to obtain the preliminary fire risk index; wherein, the expression for the preliminary fire risk index is: In the formula, This is a preliminary fire risk index. The total number of characteristic factors, For the first The weight coefficients of each feature factor, For the first The normalized values ​​of each characteristic factor.

[0079] Specifically, the weighting of feature factors is achieved using the analytic hierarchy process (AHP) combined with expert experience calibration to ensure that the weight coefficients accurately reflect the degree of influence of each feature factor on fire risk. The three-level evaluation system constructed earlier is initially adopted: the target level is the calculation of the fire risk index; the criteria level is divided into three categories: meteorological features, topographic features, and vegetation features; and the indicator level consists of specific feature factors. Experts in fire prevention and suppression are invited to conduct pairwise comparisons of the importance of indicators at each level, based on historical fire case data, to construct a judgment matrix. The rationality of the judgment matrix is ​​verified through a consistency test. If the test fails, the expert judgment results are readjusted until the consistency requirement is met.

[0080] The final weight ratios of the criterion layer and the indicator layer are determined, with the sum of the weight coefficients of all feature factors being 1. Based on the weight coefficients and the normalized eigenvectors, a preliminary fire risk index is calculated using linear weighted summation. In the above expression, This is the preliminary fire risk index for a single grid point, and the range of the preliminary fire risk index for a single grid point is a specified interval. This represents the total number of feature factors, i.e., the dimension of the normalized feature vector. For the first The weight coefficients of each feature factor; For the first The normalized values ​​of each feature factor are used for calculation. During the calculation, each value in the normalized feature vector of each grid point is multiplied by the corresponding weight coefficient, and the results are accumulated to obtain the preliminary fire risk index. At the same time, the contribution ratio of each feature factor to the index is recorded to provide a basis for subsequent fire risk factor tracing.

[0081] S33. Introduce slope correction factors and aspect correction factors to perform terrain correction on the preliminary fire risk index, obtaining the corrected fire risk index; the expression for the corrected fire risk index is: In the formula, This is the revised fire risk index. This is a preliminary fire risk index. This is the slope correction factor. This is the aspect correction factor.

[0082] Optionally, terrain factors directly affect the speed and extent of fire spread, and their amplification or reduction effect on local fire risk cannot be fully reflected by weight allocation alone. Therefore, a correction factor needs to be introduced to optimize the calculation results. In the above-mentioned corrected fire risk index expression, The revised fire risk index is controlled within a specified range. If the calculated result exceeds the upper limit, the upper limit value is used; if it is lower than the lower limit, the lower limit value is used. This is a preliminary fire risk index; This is the slope correction factor; This is the aspect correction factor.

[0083] The slope correction factor is set based on the slope levels defined above and calibrated using historical fire spread rate statistics. A steeper slope results in a larger correction factor and a higher increase in the corresponding fire risk index, reflecting the reality that steep slopes experience faster fire spread and higher risk. The aspect correction factor is set based on the influence of slope direction on sunlight and humidity. South-facing slopes receive ample sunlight and have low combustible material moisture content, so the correction factor is positive. Other slope aspects have corresponding correction factors set based on sunlight and humidity conditions, with some aspects having no increase in correction. The correction factors are periodically fine-tuned based on regional climate characteristics to ensure they are adapted to the different impacts of terrain on fire risk in different seasons.

[0084] S34. Perform smoothing filtering on the corrected fire risk index to obtain the real-time fire risk index for each grid.

[0085] Optionally, smoothing filtering is used to eliminate abrupt changes in the fire hazard index between grid points, making the spatial distribution of the index more continuous and closely reflecting the gradual changes in actual fire hazard, while preserving the characteristics of core high-risk areas. A Gaussian sliding window filtering algorithm is adopted, with the grid point to be processed at the center of the window. Each grid point within the window is assigned a different weight, with the central grid point having the highest weight, and the weight decreasing sequentially towards the outer edges, with the total weight being 1. The interference of isolated outliers on the results is weakened by weighted averaging.

[0086] During the filtering process, each grid point is traversed, and the corrected fire risk index within the corresponding range is extracted with that grid point as the center. If the window exceeds the boundary of the target area, the data is supplemented by repeatedly filling the boundary grid points. Then, the weighted average is calculated according to the window weight, which serves as the real-time fire risk index for that grid point. After filtering, the results are verified by calculating the deviation value of the index before and after filtering. The deviation range is controlled to ensure that the filtering does not lose key fire risk information. At the same time, outliers that still exist after filtering are removed. The data is supplemented by the mean of neighboring grid points, and finally, a spatially continuous and accurate real-time fire risk index for each grid is obtained, providing a reliable input for subsequent level mapping.

[0087] In an optional embodiment, time-series trend prediction processing is performed based on real-time fire risk index and historical fire risk data to calculate the change in fire risk index over a specific future time period, thereby obtaining the future fire risk prediction level, including: S41. Construct a time series based on the corrected fire risk index, extract the historical fire risk change pattern for the same period, and obtain historical trend characteristic data.

[0088] Optionally, the time series construction uses the corrected fire risk index as the core data source, collecting data from all grid points in the target area on an hourly basis to form a continuous time series. The series length needs to cover the fire risk variation patterns under different seasons and climatic conditions to ensure the comprehensiveness of historical trend extraction. Time series data preprocessing requires removing extreme outliers, using linear interpolation to fill in missing data periods, and then smoothing the time series curve through moving average filtering to reduce the interference of short-term random fluctuations on trend extraction.

[0089] Historical fire risk variation patterns were extracted using a combination of time-series clustering and trend fitting. The data was divided into time-series intervals based on natural months and solar terms, with time-series data from the corresponding historical period used as analysis samples for each interval. Clustering algorithms were used to classify the data into different fire risk trend types, with the number of clusters determined by silhouette coefficient verification. Multinomial fitting was then applied to each type of data to obtain the variation curves and characteristic parameters of the fire risk index for the same period. These characteristic parameters include the daily average rate of change, peak occurrence time, fluctuation amplitude, and trend inflection points. These parameters were then organized into gridded data to form historical trend feature data. The core of the historical time-series pattern value represents the average fire risk index and its contribution to the trend at future points in the historical time-series, providing historical pattern support for prediction.

[0090] S42. Extract short-term weather forecast elements from meteorological characteristic data to obtain weather forecast data for a future preset time period, and obtain forecast characteristic data.

[0091] Optionally, the future preset time period is divided into short-term hourly forecasts and medium-term daily forecasts. Meteorological forecast data is obtained by connecting to the meteorological department's gridded forecast interface, and the data resolution matches the grid resolution of the target area to ensure spatial alignment. The extracted forecast elements must correspond one-to-one with the meteorological characteristic data mentioned above, including forecast values ​​for temperature, relative humidity, wind speed, wind direction, and precipitation, while also supplementing extreme weather warning indicators.

[0092] Forecast feature data processing requires initial spatiotemporal calibration, converting the coordinates of the forecast data to a coordinate system consistent with the target area. Forecast values ​​are assigned by grid point, and spatial interpolation is used to complete forecast data that crosses grid points. Subsequently, the forecast values ​​are mapped to corresponding intervals using the same Min-Max standardization method as described earlier. Then, the weights of each forecast element are determined based on the analytic hierarchy process (AHP), maintaining consistency with the meteorological feature weights. The weighted summation is used to calculate the forecast feature impact value for future times. The elements with the highest weight are wind speed and temperature forecast values, ensuring the dominant role of core meteorological elements in fire risk prediction. Simultaneously, a forecast error correction mechanism is established. Based on a comparison of past forecast values ​​and measured values, the forecast error coefficients of each element are calculated to correct the current forecast values, ensuring the reliability of the forecast data.

[0093] S43. Based on historical trend characteristic data and forecast characteristic data, a prediction index for future time is calculated; the expression for the prediction index for future time is: In the formula, For the future The fire risk index is predicted at any time. For the current moment The revised fire risk index, For the future The influence value of the forecast characteristics at any given time. This is a historical value for the same period. , , For dynamic adjustment coefficients, and .

[0094] Optionally, a multi-factor weighted fusion model is used for prediction calculation, taking into account the current basic fire risk status, short-term meteorological changes, and historical trend patterns. In the above expression for predicting the fire risk index, For the future The predicted fire risk index at any given time is controlled within a specified range; if it exceeds the range, it is treated as a boundary value. For the current moment The revised fire risk index reflects the current basic fire risk status; For the future The forecast characteristic impact value at any given time characterizes the impact of short-term weather changes on fire risk; The historical data represents the pattern value for the same period, indicating the reference value of historical fire risk patterns for current forecasts. , , The coefficient is dynamically adjusted and satisfies the following conditions: This is used to adapt the influence weights of each factor in different scenarios.

[0095] The dynamic adjustment coefficients employ an adaptive allocation strategy, dynamically optimized based on real-time scenarios. During the day, temperature and wind speed have a significant impact, and corresponding coefficient weights are set accordingly. Nighttime weather conditions are relatively stable, and historical trends are more valuable for reference, so the adjustment coefficient weights are primarily based on historical trends. During extreme weather warning periods, the weight of weather forecasts increases, and the coefficient weights are adjusted accordingly. Coefficients are periodically fine-tuned based on prediction errors, adjusting coefficient weights according to the error direction to ensure model prediction accuracy. During calculation, each grid point is substituted into the formula to calculate the predicted index for each future time point, forming gridded prediction index time-series data. Simultaneously, the contribution percentage of each coefficient to the prediction result is recorded, providing a basis for tracing the source of prediction errors.

[0096] S44. Based on the preset fire risk level threshold, determine the level of the future time prediction index, determine the fire risk level for the future period, and obtain the future fire risk prediction level.

[0097] Optionally, the preset fire risk level threshold is consistent with the fire risk level classification standard to maintain a unified judgment standard and ensure the consistency between the fire risk assessment level and the future prediction level. The level determination adopts a grid-point, time-by-time mapping method, which maps the predicted fire risk index for the short term on an hourly basis and the medium term on a daily basis to specific fire risk levels, forming time-series data on the future fire risk prediction level.

[0098] During the assessment process, the credibility of the prediction results at each grid point and at each time point is calculated simultaneously. Credibility is based on a weighted average of historical prediction accuracy and current weather forecast accuracy. Prediction results meeting the credibility standard are marked as high credibility and given priority for prevention and control; those with medium credibility are marked as moderately credible and require dynamic verification based on real-time monitoring data; and those with low credibility require manual review and adjustment. Finally, the prediction levels are summarized according to administrative divisions and time dimensions to form future fire risk prediction levels, clarifying the distribution and credibility of fire risk levels in different time periods and regions, providing precise temporal guidance for subsequent high-risk area identification and prevention measures development.

[0099] In an optional embodiment, high-risk fire areas are identified from fire hazard assessment data, and fire hazard analysis is performed on these high-risk areas to obtain fire hazard analysis results, including the following steps: S51. Extract grids of fire risk levels exceeding a preset threshold from fire risk assessment data, and perform spatial aggregation based on administrative boundaries to obtain candidate high-risk areas.

[0100] Optionally, the preset threshold is consistent with the fire risk level classification standard mentioned above, explicitly defining grids corresponding to high-risk and above levels as over-threshold grids. Simultaneously, based on future short-term fire risk prediction levels, only over-threshold grids with predicted high or above levels are retained, excluding areas where short-term fire risks can be naturally mitigated, ensuring the timeliness and urgency of the candidate areas. Administrative division boundary data uses vector boundary data, containing multi-level boundary information, and is spatially correlated with the over-threshold grids using GIS spatial overlay analysis tools.

[0101] The spatial aggregation process employs a two-way overlay strategy of grid and administrative division. First, grids exceeding the threshold are initially aggregated according to the boundaries of basic-level administrative divisions, generating continuous grid clusters within the region. Then, adjacent grid clusters crossing administrative divisions are merged to avoid administrative boundaries fragmenting the complete risk area. During aggregation, area filtering conditions are set to remove scattered grid clusters with excessively small areas, as these areas have minimal fire risk impact and do not require priority analysis, reducing unnecessary computation. After aggregation, spatial topology checks correct for issues such as overlapping boundaries and gaps. Each candidate high-risk area is assigned a unique code, and core parameters for each area are statistically correlated, including area, number of involved grid points, average fire risk index, and the proportion of extremely high-risk grids, forming a candidate high-risk area dataset, laying the foundation for subsequent ranking analysis.

[0102] S52. Statistically sort the fire risk index of the candidate high-risk areas, rank the candidate high-risk areas according to the severity of fire risk, and obtain the risk area data.

[0103] Optionally, the statistical analysis employs a multi-indicator weighted scoring method, selecting core indicators from three dimensions: fire risk intensity, impact range, and diffusion potential, to ensure that the ranking comprehensively reflects the severity of fire risk. Core statistical indicators include the regional average fire risk index, the proportion of extremely high-risk grids, the regional maximum fire risk index, and the predicted short-term increase in the fire risk index, with each indicator assigned a corresponding weight.

[0104] After statistical calculations, candidate high-risk areas are ranked from highest to lowest based on their comprehensive scores. Areas with the same score are ranked according to the percentage of extremely high-risk grids, with areas having a higher percentage ranked higher. The ranking results are divided into three priorities, each corresponding to a different prevention and control response level. The final risk area data includes complete ranking information, basic regional attributes, multi-dimensional statistical indicators, priority indicators, and predictions of future fire risk trends. The data is stored in structured tables and linked to corresponding gridded spatial data, supporting integrated display with GIS maps for easy and rapid location of core risk areas.

[0105] S53. Perform environmental overlay analysis on the risk area data to obtain risk distribution data.

[0106] Optionally, environmental overlay analysis employs GIS spatial analysis technology to overlay risk area data with multi-dimensional environmental element data into layers, accurately depicting the correlation between risk areas and environmental elements, and providing a basis for risk assessment and measure formulation. The overlaid environmental element data covers four core categories: topography, vegetation, meteorology, and human facilities, all maintaining the same grid resolution as the target area to ensure overlay accuracy. Topographic elements are overlaid with previously generated slope, aspect, and special hazardous terrain layers to analyze the topographic distribution characteristics of areas with different risk priorities; vegetation elements are overlaid with previously generated vegetation type, vegetation load, and litter thickness layers to clarify the dominant vegetation types and the distribution range of flammable vegetation within the risk area.

[0107] Meteorological elements are overlaid with real-time meteorological monitoring data and short-term forecast data layers, including parameters such as wind speed, wind direction, temperature, and humidity, to analyze the impact of meteorological conditions on the spread of fire risk in high-risk areas. Combined with wind direction data, the potential direction of fire spread and the environmental areas involved are predicted. Human infrastructure elements are overlaid with vector layers of residential areas, key protected targets, roads, and fire stations to statistically analyze the distribution, quantity, and distance of facilities within and around the high-risk area, assessing the threat level of fire to human infrastructure and the accessibility of emergency rescue. After the overlay analysis is completed, risk distribution data is generated, the core of which includes a refined risk distribution layer and an attribute association table. Layers are labeled with different color depths according to priority and support linkage switching with auxiliary layers such as terrain, vegetation, and facilities. The attribute table contains environmental element association information for each high-risk area, achieving spatial visualization and attribute traceability.

[0108] S54. Perform association rule matching on the risk distribution data, query the preset risk prevention knowledge base, generate risk prevention measures suggestions including suggestions for setting up isolation zones and resource allocation, integrate and generate a fire risk assessment report, and obtain fire risk analysis results.

[0109] Optionally, the pre-built risk prevention knowledge base is constructed using a structured database, storing content covering three main categories: historical fire response cases, expert experience solutions, and industry prevention and control standards. It supports rule matching based on combinations of environmental elements and risk priorities. The knowledge base construction must consider the characteristics of the regional forest and grassland areas, recording the handling process, prevention and control measures, and effectiveness evaluations of local fire cases. Experts in fire prevention and suppression should be invited to calibrate the relevance of the recommended measures, while also aligning with industry prevention and control standards to ensure compliance. The knowledge base is regularly updated based on new cases and technical standards to maintain its timeliness.

[0110] Association rule matching employs a corresponding algorithm, using combinations of environmental elements and risk priorities from the risk distribution data as retrieval conditions to match the optimal prevention and control plan from the knowledge base. For recommendations on establishing firebreaks, parameters are set based on the matching results, and the width of the firebreak is determined in conjunction with the slope. A combination of mechanical clearing and manual trimming is used to clear debris to the corresponding thickness. The direction of the firebreak is set in conjunction with contour lines and prevailing wind direction to reduce the risk of fire crossing. Resource allocation recommendations are generated based on emergency rescue accessibility and risk priority. High-priority risk areas prioritize the allocation of specialized firefighting equipment and teams to ensure rapid arrival at the core area; other priority areas allocate conventional firefighting teams and emergency supplies, specifying the specifications and quantities of supplies allocated.

[0111] After the recommended measures are generated, risk area data, risk distribution data, and the recommended measures are integrated to generate a standardized fire risk assessment report. The report includes four core components: a risk overview, risk distribution charts, tiered prevention measures, and an emergency resource allocation plan. The report generation efficiency must meet real-time prevention and control needs, support integration with subsequent early warning result modules, and reserve interfaces for manual revisions and additions. Ultimately, a complete fire risk analysis result is formed, providing comprehensive support for early warning information display and prevention and control decision-making.

[0112] In an optional embodiment, a fire risk warning result is generated based on the fire risk analysis results, including the following steps: S61. Based on the risk area data in the fire risk analysis results, determine the scope and target of the early warning information to be pushed, and obtain the target push list.

[0113] Optionally, the scope of the alert can be determined using a dual strategy of "geographical scope + risk-related scope". The geographical scope is based on the administrative boundaries in the risk area data, accurately covering townships and villages involved in high-risk areas at all levels, while extending outwards to establish a buffer zone. Only advisory alerts are sent within the buffer zone, while mandatory alerts are sent to core risk areas. The risk-related scope is determined in conjunction with the emergency rescue chain, covering fire stations, emergency material reserve points, and grassroots governance units within and around the risk area, ensuring that relevant rescue entities receive information first.

[0114] The system categorizes and defines the recipients of information based on their authority and responsibilities, forming a multi-level notification system. The administrative level includes heads and staff of emergency management departments at all levels, with notification permissions allocated according to risk area priority. Different priority risk areas are then sent to the corresponding management levels. The grassroots execution level includes township officials, village grid workers, forest rangers, and firefighters involved in the risk areas, with information on the corresponding risk areas precisely pushed according to their jurisdiction. The public level sends evacuation advisories to residents, tourists, and workers within the risk areas and buffer zones.

[0115] Based on the classification results, a target push list is generated. The list is stored in a structured format and includes information such as the name, organization, contact information, permission level, corresponding risk area code, and push channel preference of the target. At the same time, the information receiving priority is marked to ensure that core targets receive the information first.

[0116] S62. Based on different risk levels and push recipient permissions, match the corresponding warning template and generate warning information.

[0117] Optionally, the preset early warning template library is built according to the two dimensions of risk level + object permission. The template library is stored on an encrypted server and is regularly updated and optimized in combination with industry standards and handling experience. It covers three major types: instruction type, analysis type, and prompt type, to adapt to the information needs of different push objects.

[0118] The instruction templates are designed for grassroots execution levels and fire brigades, specifying handling requirements for different risk levels. High-priority risk area templates include mandatory requirements such as immediate activation of emergency response, team assembly requirements, and instructions for setting up isolation zones; medium-priority templates include control requirements such as increased patrol frequency and pre-deployment of supplies; low-priority templates include reminders such as routine patrols and hazard identification.

[0119] The assessment templates are designed for administrative management levels, presenting content differently according to authority levels. High-level templates focus on the distribution of high-risk areas across the entire region, overall situation assessment, and cross-regional resource allocation suggestions; medium-level templates focus on details of risk areas within the jurisdiction and joint contingency plans with adjacent areas; low-level templates focus on specific risk area handling plans and local resource allocation lists.

[0120] The alert templates are designed for the general public, using plain language to clearly indicate evacuation routes, temporary shelter locations, and emergency contact information, avoiding technical jargon. They are also optimized for different push channels. SMS templates are limited to a reasonable character count, app pop-up templates can include thumbnails of risk distribution maps, and broadcast templates use conversational language for easy dissemination. After matching, core information such as risk area data and preventative measures suggestions are populated into the template to generate personalized alert messages. The information is also encrypted to ensure secure transmission.

[0121] S63. Based on the target push list, the early warning information is simultaneously pushed to the terminal equipment in the affected area to obtain the fire risk early warning result.

[0122] Optionally, the push notification employs a multi-channel collaborative push strategy, allocating push channels according to channel preferences and target audience types in the target push list: Administrative and grassroots execution levels simultaneously push information through three channels—encrypted office apps, SMS, and pop-up windows on office computers—ensuring timely delivery; fire brigades additionally push voice commands via dedicated walkie-talkie clusters, adapting to field operation scenarios; and the general public pushes information through village broadcasts, public area displays, civilian apps, and SMS, achieving full coverage without blind spots. The push process relies on the aforementioned TCP / UDP protocol encrypted channel, connecting to the interfaces of various terminal devices to ensure information transmission is leak-free and delay-free.

[0123] After the push is completed, a dual verification mechanism is activated. The delivery status is confirmed by the receipt information from the terminal device. For those not delivered, a backup channel is used for re-push, with a reasonable re-push frequency set. If multiple pushes fail to deliver consecutively, feedback is sent to the corresponding level of emergency management department for manual confirmation. Simultaneously, all push process data is recorded, including push time, channel, reception status, and feedback results, forming a push log and archiving it together with the fire risk warning result to support subsequent traceability and review. The final generated fire risk warning result includes three main modules: personalized warning information, push log, and terminal display adaptation data. It can directly connect to preset terminal devices to achieve visualized display of warning information and instruction transmission, providing real-time support for fire prevention and control.

[0124] The aforementioned intelligent early warning method for forest and grassland fire risk integrates multi-dimensional heterogeneous data such as meteorology, combustibles, topography, and vegetation to achieve comprehensive feature extraction and deep fusion of fire risk influencing factors. Based on a refined grid and dynamic weight model, it generates a real-time fire risk index that better reflects the actual geographical environment. Combined with historical patterns and weather forecasts, it achieves dynamic prediction of future changes in fire risk levels. Furthermore, it automatically identifies and sorts high-risk areas, generating an analysis report that includes a risk distribution map and specific prevention measures recommendations. Ultimately, it achieves accurate, rapid, and tiered delivery of early warning information, enhancing the comprehensiveness, accuracy, foresight, and action guidance of forest and grassland fire risk early warning, and providing strong support for scientific fire prevention decision-making.

[0125] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0126] Based on the same inventive concept, this application also provides a forest and grassland fire risk intelligent early warning system for implementing the above-mentioned intelligent early warning method for forest and grassland fire risk. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the intelligent early warning system for forest and grassland fire risk provided below can be found in the limitations of the intelligent early warning method for forest and grassland fire risk described above, and will not be repeated here.

[0127] In one exemplary embodiment, such as Figure 2 As shown, a schematic diagram of a forest and grassland fire risk intelligent early warning system 10 is provided, including: The multi-source data acquisition module 11 is used to acquire multi-source heterogeneous data of forest and grassland in the target area; among which, the multi-source heterogeneous data includes meteorological data and combustible moisture content detection data; The fire risk feature extraction module 12 is used to extract fire risk factor features from multi-source heterogeneous data to obtain fire risk feature data; the fire risk feature data includes meteorological feature data, terrain feature data and vegetation feature data; The fire risk level assessment module 13 is used to assess the fire risk characteristics to obtain fire risk assessment data; the fire risk assessment data includes the fire risk assessment level and the future fire risk prediction level. The fire risk analysis module 14 is used to identify high-risk fire areas from fire risk assessment data, and to perform fire risk analysis on high-risk fire areas to obtain fire risk analysis results; the fire risk analysis results include risk area data, risk distribution data, and risk prevention measures recommendations; The fire hazard warning module 15 is used to generate fire hazard warning results based on the fire hazard analysis results; the fire hazard warning results are used to instruct preset equipment to display the fire hazard analysis results.

[0128] Furthermore, the fire hazard feature extraction module 12 is also used for: S11. Perform meteorological feature extraction processing on meteorological data in multi-source heterogeneous data, calculate the time change rate of temperature, relative humidity, wind speed, wind direction and precipitation, and obtain meteorological feature data. S12. Based on the digital elevation model data, perform terrain factor calculation and processing, extract slope, aspect and altitude information to obtain terrain feature data; S13. Based on remote sensing image data, extract vegetation distribution and vegetation type information, and combine combustible moisture content detection data to calculate vegetation load and litter layer thickness to obtain vegetation characteristic data. S14. Perform multi-dimensional feature fusion processing on meteorological feature data, terrain feature data and vegetation feature data to obtain fire risk feature data.

[0129] Furthermore, the fire hazard assessment module 13 is also used for: S21. Based on the fire hazard feature data, perform gridding processing to divide the target area into grid points of preset resolution to obtain gridded fire hazard feature data; S22. Perform weighted calculation on the gridded fire risk characteristic data to obtain the real-time fire risk index for each grid. S23. Map the real-time fire risk index to a level. Based on the preset fire risk level classification standard, map the value to the corresponding fire risk level to obtain the fire risk assessment level. S24. Based on real-time fire risk index and historical fire risk data, perform time-series trend prediction processing to calculate the change of fire risk index in a specific future time period and obtain the future fire risk prediction level.

[0130] Furthermore, the fire hazard assessment module 13 is also used for: S31. Normalize the gridded fire hazard feature data to obtain the normalized feature vector; S32. Assign feature factor weights to the normalized feature vector to obtain the weight coefficients of each feature factor. Based on the weight coefficients of each feature factor, perform a weighted summation of the feature factors to obtain the preliminary fire risk index; wherein, the expression for the preliminary fire risk index is: In the formula, This is a preliminary fire risk index. The total number of characteristic factors, For the first The weight coefficients of each feature factor, For the first Normalized values ​​of each feature factor; S33. Introduce slope correction factors and aspect correction factors to perform terrain correction on the preliminary fire risk index, obtaining the corrected fire risk index; wherein, the expression for the corrected fire risk index is: In the formula, This is the revised fire risk index. This is a preliminary fire risk index. This is the slope correction factor. This is the aspect correction factor; S34. Perform smoothing filtering on the corrected fire risk index to obtain the real-time fire risk index for each grid.

[0131] Furthermore, the fire hazard assessment module 13 is also used for: S41. Construct a time series based on the corrected fire risk index, extract the historical fire risk change pattern for the same period, and obtain historical trend characteristic data; S42. Extract short-term weather forecast elements from meteorological characteristic data to obtain weather forecast data for a future preset time period and obtain forecast characteristic data. S43. Based on historical trend characteristic data and forecast characteristic data, a prediction index for future time is obtained through prediction calculation; the expression for the prediction index for future time is: In the formula, For the future The fire risk index is predicted at any time. For the current moment The revised fire risk index, For the future The influence value of the forecast characteristics at any given time. This is a historical value for the same period. , , For dynamic adjustment coefficients, and ; S44. Based on the preset fire risk level threshold, determine the level of the future time prediction index, determine the fire risk level for the future period, and obtain the future fire risk prediction level.

[0132] Furthermore, the fire hazard analysis module 14 is also used for: S51. Extract grids of fire risk levels exceeding a preset threshold from fire risk assessment data, and perform spatial aggregation based on administrative boundaries to obtain candidate high-risk areas. S52. Statistically sort the fire risk index of the candidate high-risk areas, rank the candidate high-risk areas according to the severity of fire risk, and obtain the risk area data. S53. Perform environmental overlay analysis on the risk area data to obtain risk distribution data; S54. Perform association rule matching on the risk distribution data, query the preset risk prevention knowledge base, generate risk prevention measures suggestions including suggestions for setting up isolation zones and resource allocation, integrate and generate a fire risk assessment report, and obtain fire risk analysis results.

[0133] Furthermore, the fire hazard warning module 15 is also used for: S61. Based on the risk area data in the fire risk analysis results, determine the scope and target of the early warning information to be pushed, and obtain the target push list; S62. Based on different risk levels and push recipient permissions, match the corresponding warning template and generate warning information; S63. Based on the target push list, the early warning information is simultaneously pushed to the terminal equipment in the affected area to obtain the fire risk early warning result.

[0134] In one embodiment, such as Figure 3 A computer device 300 is provided, comprising: At least one processor 301, and at least one memory 302 communicatively connected to said processor 301; said memory stores application code executable by said processor, said application code being executed by said processor to enable said processor to perform the steps of the intelligent early warning method for forest and grassland fire risk as described above; The computer device may also include: sensor 303; The processor 301, memory 302, and sensor 303 can be connected via bus 304 or other means. The figure shows an example of connection via bus 304. Figure 3 The character is represented by a single thick line, but this does not mean that there is only one bus or a type of bus.

[0135] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0136] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0137] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for intelligent early warning of forest and grassland fire risk, characterized in that, The method includes: S1. Acquire multi-source heterogeneous data of forests and grasslands in the target area; wherein, the multi-source heterogeneous data includes meteorological data and combustible material moisture content detection data; S2. Extract fire risk factor features from the multi-source heterogeneous data to obtain fire risk feature data; the fire risk feature data includes meteorological feature data, terrain feature data, and vegetation feature data; S3. Perform a fire risk assessment on the fire risk characteristic data to obtain fire risk assessment data; the fire risk assessment data includes the fire risk assessment level and the future fire risk prediction level. S4. Identify high-risk fire areas from the fire risk assessment data, and perform fire risk analysis on the high-risk fire areas to obtain fire risk analysis results; the fire risk analysis results include risk area data, risk distribution data, and risk prevention measures recommendations; S5. Generate a fire risk warning result based on the fire risk analysis result; the fire risk warning result is used to instruct preset equipment to display the fire risk analysis result.

2. The method according to claim 1, characterized in that, The step of extracting fire hazard factor features from the multi-source heterogeneous data to obtain fire hazard feature data includes: S11. Perform meteorological feature extraction processing on the meteorological data in the multi-source heterogeneous data, calculate the time change rate of temperature, relative humidity, wind speed, wind direction and precipitation, and obtain meteorological feature data; S12. Based on the digital elevation model data, perform terrain factor calculation and processing, extract slope, aspect and altitude information to obtain terrain feature data; S13. Extract vegetation distribution and vegetation type information based on remote sensing image data, and calculate vegetation load and litter layer thickness by combining the combustible moisture content detection data to obtain vegetation characteristic data. S14. Perform multi-dimensional feature fusion processing on the meteorological feature data, the terrain feature data, and the vegetation feature data to obtain the fire risk feature data.

3. The method according to claim 2, characterized in that, The fire risk assessment of the fire risk characteristic data to obtain fire risk assessment data includes: S21. Based on the fire hazard feature data, perform gridding processing to divide the target area into a grid of points with a preset resolution to obtain gridded fire hazard feature data; S22. Perform weighted calculation on the gridded fire risk feature data to obtain the real-time fire risk index of each grid. S23. Map the real-time fire risk index to a level. Based on the preset fire risk level classification standard, map the value to the corresponding fire risk level to obtain the fire risk assessment level. S24. Based on the real-time fire risk index and historical fire risk data, perform time-series trend prediction processing to calculate the change in the fire risk index over a specific future time period, and obtain the future fire risk prediction level.

4. The method according to claim 3, characterized in that, The weighted calculation of the gridded fire hazard feature data to obtain the real-time fire hazard index for each grid includes: S31. Normalize the gridded fire hazard feature data to obtain a normalized feature vector; S32. Assign feature factor weights to the normalized feature vector to obtain the weight coefficients of each feature factor. Based on the weight coefficients of each feature factor, perform a weighted summation of the feature factors to obtain a preliminary fire risk index; wherein, the expression for the preliminary fire risk index is: In the formula, This is a preliminary fire risk index. The total number of characteristic factors, For the first The weight coefficients of each feature factor, For the first Normalized values ​​of each feature factor; S33. Introduce slope correction factors and aspect correction factors to perform terrain correction on the preliminary fire risk index to obtain the corrected fire risk index; wherein, the expression for the corrected fire risk index is: In the formula, This is the revised fire risk index. This is a preliminary fire risk index. This is the slope correction factor. This is the aspect correction factor; S34. Perform smoothing filtering on the corrected fire risk index to obtain the real-time fire risk index of each grid.

5. The method according to claim 4, characterized in that, The step of performing time-series trend prediction processing based on the real-time fire risk index and historical fire risk data to calculate the change in the fire risk index over a specific future time period and obtain the future fire risk prediction level includes: S41. Construct a time series based on the modified fire risk index, extract the historical fire risk change pattern for the same period, and obtain historical trend feature data; S42. Perform short-term weather forecast element extraction processing on the meteorological feature data to obtain weather forecast data for a future preset time period, and obtain forecast feature data; S43. Based on the historical trend feature data and the forecast feature data, a prediction calculation is performed to obtain the future time prediction index; wherein, the expression for the future time prediction index is: In the formula, For the future The fire risk index is predicted at any time. For the current moment The revised fire risk index, For the future The influence value of the forecast characteristics at any given time. This is a historical value for the same period. , , For dynamic adjustment coefficients, and ; S44. Based on the preset fire risk level threshold, the future time prediction index is judged to determine the fire risk level for the future time period, and the future fire risk prediction level is obtained.

6. The method according to claim 1, characterized in that, The process of identifying high-risk fire areas from the fire risk assessment data and performing fire risk analysis on these high-risk areas to obtain fire risk analysis results includes: S51. Extract grids of fire risk levels exceeding a preset threshold from the fire risk assessment data, and perform spatial aggregation based on administrative division boundaries to obtain candidate high-risk areas. S52. Statistically sort the fire risk index of the candidate high-risk areas, and rank the candidate high-risk areas according to the severity of fire risk to obtain risk area data; S53. Perform environmental overlay analysis on the risk area data to obtain risk distribution data; S54. Perform association rule matching on the risk distribution data, query the preset risk prevention knowledge base, generate risk prevention measures suggestions including suggestions for setting up isolation zones and resource allocation, integrate and generate a fire risk assessment report, and obtain the fire risk analysis results.

7. The method according to claim 1, characterized in that, The step of generating a fire risk warning result based on the fire risk analysis results includes: S61. Based on the risk area data in the fire risk analysis results, determine the scope and target of the early warning information to be pushed, and obtain the target push list; S62. Match the corresponding warning template and generate warning information according to different risk levels and push object permissions; S63. Based on the target push list, the early warning information is synchronously pushed to the terminal devices in the affected area to obtain the fire risk early warning result.

8. A forest and grassland fire risk intelligent early warning system, characterized in that, The system includes: A multi-source data acquisition module is used to acquire multi-source heterogeneous data of forests and grasslands in the target area; wherein, the multi-source heterogeneous data includes meteorological data and combustible material moisture content detection data; The fire risk feature extraction module is used to extract fire risk factor features from the multi-source heterogeneous data to obtain fire risk feature data; the fire risk feature data includes meteorological feature data, terrain feature data and vegetation feature data. The fire risk level assessment module is used to assess the fire risk characteristics to obtain fire risk assessment data; the fire risk assessment data includes the fire risk assessment level and the future fire risk prediction level. The fire risk analysis module is used to identify high-risk fire areas from the fire risk assessment data, and to perform fire risk analysis on the high-risk fire areas to obtain fire risk analysis results; the fire risk analysis results include risk area data, risk distribution data, and risk prevention measures recommendations; The fire hazard warning module is used to generate a fire hazard warning result based on the fire hazard analysis results; the fire hazard warning result is used to instruct preset devices to display the fire hazard analysis results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images