A control method for forest fire early warning system

Abstract

The present application provides a control method for a forest fire early warning system, comprising: conducting combustion characteristic tests on different types of litter, determining the ignition point, calorific value, and combustion rate parameters of the litter, and obtaining combustion characteristic data of different types of litter; analyzing the spatial distribution characteristics of the fire risk of the litter layer on the basis of the classification of fire risk levels of the litter layer, identifying high-incidence areas of fire risk, and using a spatial autocorrelation analysis method to represent the spatial aggregation and correlation of the fire risk of the litter layer; analyzing the dynamic change characteristics of the fire risk of the litter layer through the evaluation results of the fire risk of the litter layer, using a time series analysis method to summarize the seasonal change law and interannual change trend of the fire risk of the litter layer, and identifying the high-incidence period of fire risk; using a geographic detector method to quantitatively analyze the influence of terrain, meteorology, and vegetation on the fire risk of the litter layer, identifying the dominant key factors, and exploring the key factors affecting the fire risk of the litter layer.

Description

Technical Field

[0001] The invention relates to the field of information technology, and in particular to a control method for a forest fire early warning system. Background Art

[0002] Traditional satellite remote sensing has a low resolution and it is difficult to accurately obtain spatial distribution information of forest litter. Although drones have high resolution, their patrol range is limited, making it difficult to achieve large-scale continuous monitoring. In addition, the spatial distribution and cumulative characteristics of the forest litter layer are complex and changeable, and the combustion characteristics of different types of litter vary significantly. Traditional fire risk assessment methods are difficult to accurately characterize the flammability and fire risk level of the litter layer. It is urgent to carry out targeted research to reveal the spatial distribution law and dynamic change mechanism of the litter layer, clarify the differences in the combustion characteristics of different types of litter and their influencing factors, explore the key characteristic parameters of the litter layer fire risk, and construct a litter layer fire risk assessment index system. At the same time, it is necessary to develop rapid identification and extraction technology for litter layer fire risk, establish an intelligent assessment model for litter layer fire risk, and form a refined early warning method for litter layer fire risk, so as to provide accurate auxiliary decision-making basis for forest fire prevention. However, the high heterogeneity of the spatial distribution and type composition of the litter layer, the complexity and variability of the combustion process, and the dynamic variability of external environmental conditions bring many technical challenges to the extraction and assessment of fire hazard characteristics of the litter layer, which need to be overcome urgently. Summary of the invention

[0003] The present invention provides a control method for a forest fire early warning system, which mainly includes:

[0004] The spatial distribution data of litter in the area are obtained through manual inspection and reporting, drone monitoring and satellite remote sensing fire monitoring, including the coverage, thickness and density of litter. A litter spatial distribution database is established based on the obtained data to obtain the spatial distribution characteristics of litter. Combustion characteristic tests are carried out for different types of litter to determine the ignition point, calorific value and burning rate parameters of litter to obtain the combustion characteristic data of different types of litter. According to the spatial distribution characteristics and combustion characteristics of litter, a litter layer fire risk assessment index system is constructed, and the key factors affecting the fire risk, including litter load, litter thickness, litter density and terrain factors, are combined to determine the dominant key factors using the hierarchical analysis method. The constructed litter layer fire risk assessment index system is used to quantitatively assess the fire risk of the litter layer in the study area, and the fuzzy comprehensive evaluation method is used to quantify and score each assessment index to obtain the litter layer fire risk level distribution map. On the basis of the litter layer fire risk level division, the litter layer fire risk level is analyzed. The spatial distribution characteristics of the fire risk of the litter layer are used to identify the high-incidence areas of fire risk. The spatial autocorrelation analysis method is used to represent the spatial clustering and correlation of the fire risk of the litter layer. The dynamic change characteristics of the fire risk of the litter layer are analyzed through the fire risk assessment results of the litter layer. The seasonal change law and interannual change trend of the fire risk of the litter layer are summarized by the time series analysis method, and the high-incidence period of fire risk is identified. The geographic detector method is used to quantitatively analyze the influence of topography, meteorology, and vegetation on the fire risk of the litter layer, identify the dominant key factors, and explore the key factors affecting the fire risk of the litter layer. Based on the fire risk assessment results of the litter layer, fire risk management measures for the litter layer are proposed, and fire prevention measures are formulated for the high-incidence areas and high-incidence periods of fire risk. The fire prevention measures include strengthening litter cleaning, optimizing vegetation structure, and establishing fire isolation belts. A fire risk monitoring and early warning system for the litter layer is established in advance to obtain litter status information in real time. Combined with meteorological data, the support vector machine algorithm is used to dynamically warn the fire risk of the litter layer. If there is a fire risk in the litter layer, a fire risk warning information will be issued.

[0005] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0006] The present invention discloses a control method for a forest fire early warning system. The spatial distribution characteristics of litter are obtained through multi-source data, and combustion parameters of different types of litter are obtained by conducting combustion characteristic tests. A litter layer fire risk assessment index system is constructed, and the dominant factors are determined by the hierarchical analysis method. The study area is quantitatively evaluated by the fuzzy comprehensive evaluation method to obtain a fire risk level distribution map. The spatial distribution and dynamic change characteristics of fire risk are analyzed to identify high-incidence areas and time periods. The influencing factors are analyzed by the geographic detector method. Targeted fire prevention measures are proposed based on the evaluation results. A real-time monitoring and early warning system is established, and a support vector machine algorithm is used in combination with meteorological data for dynamic early warning. The present invention realizes accurate assessment, dynamic monitoring and early warning of litter layer fire risk, and provides a scientific basis and technical support for forest fire prevention. By identifying high-incidence areas and time periods of fire risk and formulating targeted prevention and control measures, the risk of forest fires can be effectively reduced and the level of forest resource protection and management can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 The present invention is a flow chart of a control method of a forest fire early warning system.

[0008] Figure 2 It is a schematic diagram of a control method of a forest fire early warning system of the present invention.

[0009] Figure 3 This is another schematic diagram of a control method for a forest fire early warning system according to the present invention. DETAILED DESCRIPTION

[0010] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0011] like Figure 1-3 In this embodiment, a control method of a forest fire early warning system may specifically include:

[0012] Step S101, obtain the spatial distribution data of litter in the area through manual inspection and reporting, drone monitoring and satellite remote sensing fire monitoring, including the coverage, thickness and density information of litter, establish a litter spatial distribution database based on the acquired data, and obtain the litter spatial distribution characteristics.

[0013] A multispectral camera mounted on an unmanned aerial vehicle is used to obtain surface images, and spectral data of visible light bands and near-infrared bands are extracted from the surface images. A standardized vegetation index is calculated based on the spectral data. A linear regression association model is established based on the standardized vegetation index and the measured surface litter thickness value, and litter distribution data is obtained from the linear regression association model. For the surface reflectance data in the satellite remote sensing image, the spectral similarity value is calculated using the spectral curves of the near-infrared band and the infrared band. If the spectral similarity value is greater than a preset threshold, it is determined to be a litter covered area. The high-resolution data collected by the unmanned aerial vehicle is subjected to principal component transformation and resampling processing, and spatial registration parameters are obtained through ground control points. The spatial correspondence between the unmanned aerial vehicle data and the satellite data is realized according to the spatial registration parameters. The vegetation index change value is calculated based on the time series remote sensing image sequence, and the seasonal change characteristic curve is extracted using the time series decomposition method. The litter accumulation period is determined based on the slope of the change characteristic curve and the inflection point position.

[0014] Specifically, the spectral data of the visible light band of 390-700 nanometers and the near-infrared band of 750-900 nanometers are obtained by collecting surface images with a multispectral camera carried by an unmanned aerial vehicle. The vegetation coverage index is obtained through the standardized vegetation index calculation formula. A linear regression relationship is established based on the vegetation coverage index and the measured surface litter thickness value, and the litter distribution data is extracted from the multispectral image. For the surface reflectance data in the 30-meter resolution satellite remote sensing image, the spectral curves of the near-infrared band and the infrared band are selected to calculate the spectral similarity, and the similarity threshold is set to 0.85 to distinguish the litter coverage area, and the litter density is divided according to the spectral similarity. For the acquired drone images and satellite remote sensing data, the principal component transformation is used to resample the drone high-resolution data, and the two types of data are spatially aligned through ground control points. The litter coverage range and thickness distribution are extracted from the spatially aligned data. The vegetation index change value in the study area was calculated based on the remote sensing image sequence obtained every 15 days. The seasonal change characteristics were extracted through time series decomposition. The litter accumulation period was determined based on the slope and inflection point characteristics of the change curve, and the dynamic change law of litter was extracted from the time series image data. According to the requirements of spectral data processing, during the process of surface image collection, the UAV flight altitude is set at 80 meters, and a multispectral camera is equipped to collect surface images. The central wavelength of the visible light band is set at 550 nanometers, the bandwidth is 40 nanometers, the central wavelength of the near-infrared band is set at 850 nanometers, the bandwidth is 60 nanometers, and the sampling interval is 0.1 meters. The surface spectral reflectance data is obtained. When the calculated vegetation index is less than 0.2, it is determined to be a litter covered area. The vegetation index between 0.2 and 0.5 is determined to be a mixed coverage area. When it is greater than 0.5, it is determined to be a vegetation covered area. A litter thickness prediction function is established through linear regression. When the spectral reflectance in the near-infrared band is less than 0.3, the litter thickness is less than 2 centimeters. When the reflectivity is between 0.3 and 0.5, the thickness is 2 to 5 centimeters. When the reflectivity is greater than 0.5, the thickness is greater than 5 centimeters. The spatial resolution of the satellite remote sensing data obtained is 30 meters. The peak wavelength of the spectral curve in the study area is selected to match the standard litter spectral curve. The spectral similarity threshold is set to 0.85. The similarity greater than 0.85 is determined as a high-density litter area, the similarity between 0.7 and 0.85 is determined as a medium-density area, and the similarity less than 0.7 is determined as a low-density area. The principal component transformation is used to resample the 0.1-meter resolution data of the drone. Five ground control points in the study area are selected to achieve two types of data registration, and the registration accuracy is better than 1 pixel. The vegetation index data in the study area are extracted from the remote sensing images obtained every 15 days, and the change curve of 12 consecutive data periods is calculated. When the slope of the curve is greater than 0.05 and continues to rise for more than 3 periods, it is determined to be the litter accumulation stage, when the slope is less than -0.05 and continues to decline for more than 3 periods, it is determined to be the litter decomposition stage, and the slope between -0.05 and 0.05 is determined to be the stable stage.

[0015] Step S102, conducting combustion characteristic tests on different types of litter, measuring the ignition point, calorific value, and combustion rate parameters of the litter, and obtaining combustion characteristic data of different types of litter.

[0016] The temperature data output by the litter surface temperature sensor and the internal temperature data output by the optical fiber temperature sensor are collected. If the temperature data rising rate exceeds the preset temperature threshold and the duration exceeds the preset time threshold, the litter ignition temperature value is obtained; according to the heat flux data output by the heat flux sensor and the surface radiation intensity data output by the infrared spectrometer, when the heat flux data is higher than the preset heat flux threshold and the duration exceeds the preset duration, the calorific value parameter is obtained; for the image sequence collected by the infrared thermal imager, several principal components of the image sequence are extracted to obtain heat source characteristic data, the combustion area boundary is divided according to the heat source characteristic data, and the combustion rate is calculated from the combustion area boundary; the oxygen concentration data, carbon monoxide concentration data, and carbon dioxide concentration data output by the gas sensor are collected, and a gas parameter data set is established according to the gas concentration data; the temperature data, the heat flux data, and the gas parameter data set are subjected to multi-parameter correction by Kalman filtering, and the combustion characteristic parameters are obtained from the corrected data.

[0017] Specifically, the surface temperature data of the fallen leaves are collected by a thermocouple temperature sensor with a range of 0-1200℃ and a sampling frequency of 10Hz, and the internal temperature changes are measured by a fiber optic temperature sensor with a range of 0-800℃ and a sampling frequency of 1Hz. When the temperature rise rate exceeds 50℃ / second and the duration exceeds 3 seconds, it is determined to be the ignition temperature value. A heat flux sensor with a detection range of 0-100kW / m² and a response time of 0.1 seconds is used to monitor the heat change data. The surface radiation intensity is obtained by an infrared spectrometer with a wavelength of 3-5 microns and a spectral resolution of 0.1 microns. The calorific value parameters are recorded when the heat flux value is stable at more than 20kW / m² for 5 seconds. For the continuous image data collected by the infrared thermal imager with a resolution of 640×480 pixels and a temperature resolution of 0.1℃, the first three principal components in the image sequence are selected to extract the heat source characteristics, and the combustion area boundary is divided by setting a temperature threshold of 300℃, and the combustion rate is calculated from the boundary expansion speed. The oxygen concentration, carbon monoxide concentration, and carbon dioxide concentration data were collected by gas sensors with a sampling frequency of 10 Hz. The temperature data, heat flux data, and gas concentration data were corrected by multi-parameters through Kalman filtering, and the combustion characteristic parameters were obtained from the data set continuously monitored for 30 minutes. During the combustion characteristic measurement process, the thermocouple temperature sensor was arranged at 16 measuring points to form a 4×4 grid array, with a measuring point spacing of 20 cm. The surface temperature data was continuously recorded at a sampling frequency of 10 Hz. When the temperature at any measuring point exceeded 150°C, the optical fiber temperature sensor was activated, and 8 measuring points were arranged inside the litter to collect temperature changes. The internal temperature data was recorded at a sampling frequency of 1 Hz. If the temperature rise rate exceeded 50°C / second and lasted for more than 3 seconds, the temperature point was recorded as the ignition point temperature. The temperature field formed by the 16 surface measuring points and the 8 internal measuring points was spatially interpolated to obtain the temperature distribution. In the heat flux monitoring, four heat flux sensors with a response time of 0.1 seconds are deployed to measure the heat change in the range of 0-100kW / m². The sensors are arranged in a square with a spacing of 40 cm. A 3-5 micron infrared spectrometer is deployed simultaneously to obtain surface radiation data. The spectral resolution is 0.1 micron and the scanning frequency is 5Hz. When the heat flux values ​​measured by the four sensors all exceed 20kW / m² and last for 5 seconds, the average heat flux is calculated as the calorific value parameter. The combustion process images are collected by a 640×480 pixel infrared thermal imager with a temperature resolution of 0.1℃ and an imaging frequency of 50Hz. The first three principal components with a variance contribution rate of 95% in the image sequence are selected as heat source features, and 300℃ is set as the combustion area boundary threshold. The boundary area changes over time to obtain the combustion rate. In gas monitoring, 4 oxygen sensors, 4 carbon monoxide sensors, and 4 carbon dioxide sensors are used. The sampling frequency is 10 Hz to record concentration data. The sensors are distributed in a square with a spacing of 20 cm. The temperature, heat flux, and gas concentration data are corrected by the Kalman filter algorithm. The filter time window is set to 1 second, and the combustion characteristic parameters are extracted from 30 minutes of continuous monitoring data.

[0018] Step S103, constructing a litter layer fire risk assessment index system based on the litter spatial distribution characteristics and combustion characteristics, combining key factors affecting fire risk, including litter load, litter thickness, litter density, and terrain factors, and using hierarchical analysis method to determine the dominant key factors.

[0019] Receive multispectral remote sensing data of near-infrared band and red light band, calculate the normalized vegetation index according to the multispectral remote sensing data, establish a regression equation through the normalized vegetation index and litter load; use the regression equation to calculate the spatial distribution value of litter load, generate an elevation map, and obtain the slope classification weight value from the elevation map; based on the litter thickness data of the region, use the spherical semivariogram and Kriging interpolation to calculate the thickness spatial distribution value, where the spherical semivariogram is γ(h)=C0+C, where C0 is the nugget value, C is the base value, and h is the sampling interval; calculate the unit area weight density value according to the thickness spatial distribution value, and the density value is obtained by the weight calculation formula; construct a judgment matrix for the load value, density value, thickness value and terrain weight value, if the consistency ratio is less than the preset threshold, it is determined that the weight calculation result is reasonable, and obtain the key factor evaluation index from the judgment matrix.

[0020] Specifically, the normalized vegetation index is calculated based on the data of the near-infrared band 810-890 nanometers and the red light band 630-690 nanometers in the multispectral remote sensing image. The exponential regression equation is established through the vegetation index and the measured litter load. The load spatial distribution value is obtained from the regression equation, and the load size is judged to vary within the range of 0.5-2.5 kilograms per square meter. The digital elevation map is generated using elevation data, and the terrain undulations of the study area are obtained through slope calculation. According to the graded weighting coefficients of 0-5 degrees, 5-15 degrees, 15-25 degrees, 25-35 degrees, and above 35 degrees, 0.2, 0.4, 0.6, 0.8, and 1.0, the terrain weight value is obtained from the slope classification. According to the measured litter thickness data in the study area, the spatial continuous distribution was calculated using the spherical semivariogram, a search radius of 300 meters, and a minimum number of data points of 12 Kriging interpolation. The density value was calculated using the unit area weight formula, and the density value was judged to vary in the range of 0.02-0.15 grams per cubic centimeter. The four key factors of load, thickness, density, and terrain were scored on a scale of 1-9 according to the 5×5 judgment matrix structure, and the weights were obtained by eigenvector calculation. If the consistency ratio was less than 0.1, the result was judged to be reasonable, and the dominant key factors were obtained from the maximum weight value. The load weight of 0.4, the density weight of 0.3, the thickness weight of 0.2, and the terrain weight of 0.1 were used as evaluation indicators. In the construction of litter fire risk assessment indicators, multispectral image data was obtained through the 30-meter resolution Landsat 8 satellite, and the near-infrared band 810-890 nanometers and the red light band 630-690 nanometers were selected to calculate the normalized vegetation index, and the vegetation index and litter load regression equation y=0.5e^(2.3x) was established, where x is the vegetation index and y is the load value, and the load distribution range is 0.5-2.5 kilograms per square meter. At the same time, slope data was extracted from the digital elevation map, and the study area was divided into five levels of terrain units. The area with a slope of 0-5 degrees accounted for 23% and was weighted 0.2, the area with a slope of 5-15 degrees accounted for 35% and was weighted 0.4, the area with a slope of 15-25 degrees accounted for 25% and was weighted 0.6, the area with a slope of 25-35 degrees accounted for 12% and was weighted 0.8, and the area with a slope of more than 35 degrees accounted for 5% and was weighted 1.0. In spatial interpolation, the spherical semivariogram function γ(h)=C0+C[1.5(h / a)-0.5(h / a)^3] is used, where C0 is the nugget value of 0.05, C is the base value of 0.95, a is the range value of 300 meters, and h is the sampling interval. The 12 nearest neighbor points are selected for Kriging interpolation to obtain the thickness distribution data. The density is calculated by the weight per unit area, and the density values ​​are distributed in the range of 0.02-0.15 grams per cubic centimeter.In the calculation of key factor weights, a 5×5 judgment matrix structure is adopted. The importance ratio of load to density is 3, the ratio of load to thickness is 5, the ratio of load to terrain is 7, the ratio of density to thickness is 3, the ratio of density to terrain is 5, and the ratio of thickness to terrain is 3. The characteristic vector calculation results in load weight of 0.4, density weight of 0.3, thickness weight of 0.2, and terrain weight of 0.1, and the consistency ratio is 0.085.

[0021] Step S104, using the constructed litter layer fire risk assessment index system, quantitatively assess the litter layer fire risk in the study area, and using a fuzzy comprehensive evaluation method to quantify and score each assessment index to obtain a litter layer fire risk level distribution map.

[0022] A load interval score is obtained according to the litter load value, the load interval score is determined by the corresponding relationship of the litter load value within the preset interval, and is multiplied by the load weight coefficient to obtain the load score; a thickness interval score is obtained for the litter thickness value, the thickness interval score is determined by the corresponding relationship of the litter thickness value within the preset interval, and is multiplied by the thickness weight coefficient to obtain the thickness score; at the same time, a density interval score is obtained for the litter density value, the density interval score is determined by the corresponding relationship of the litter density value within the preset interval, and is multiplied by the density weight coefficient to obtain the density score; a slope interval score is obtained according to the terrain slope value, the slope interval score is determined by the corresponding relationship of the terrain slope value within the preset interval, and is multiplied by the slope weight coefficient to obtain the terrain score; a quadratic curve function is used to calculate the slope interval score. Calculate the membership values ​​of the load score, the density score, the thickness score and the terrain score, the membership value is obtained by dividing the square of the interval score minus the minimum value by the square of the maximum value minus the minimum value; obtain the comprehensive score by weighted sum of the load score, the density score, the thickness score and the terrain score, if the comprehensive score is less than or equal to the first threshold, it is determined to be a low fire risk level; if the comprehensive score is greater than the first threshold and less than or equal to the second threshold, it is determined to be a medium-low fire risk level; if the comprehensive score is greater than the second threshold and less than or equal to the third threshold, it is determined to be a medium fire risk level; if the comprehensive score is greater than the third threshold and less than or equal to the fourth threshold, it is determined to be a medium-high fire risk level; if the comprehensive score is greater than the fourth threshold, it is determined to be a high fire risk level.

[0023] Specifically, a quantitative scoring standard is established based on the litter load value, and the load quantification value is obtained by assigning 20 points to a load in the range of 0-0.5 kg per square meter, 40 points to a load in the range of 0.5-1.5 kg per square meter, 60 points to a load in the range of 1.5-2.0 kg per square meter, 80 points to a load in the range of 2.0-2.5 kg per square meter, and 100 points to a load greater than 2.5 kg per square meter. The load score is calculated by multiplying the load by the weight coefficient 0.4. The thickness and density data of litter were scored separately, with 20 points assigned to the thickness in the range of 0-2 cm, 40 points assigned to the thickness in the range of 2-5 cm, 60 points assigned to the thickness in the range of 5-10 cm, 80 points assigned to the thickness in the range of 10-15 cm, and 100 points assigned to the thickness greater than 15 cm, to obtain the thickness quantification value multiplied by a weight of 0.2; the density was assigned 20 points in the range of 0-0.02 grams per cubic centimeter, 40 points in the range of 0.02-0.05 grams per cubic centimeter, 60 points in the range of 0.05-0.10 grams per cubic centimeter, 80 points in the range of 0.10-0.15 grams per cubic centimeter, and 100 points in the range of greater than 0.15 grams per cubic centimeter, to obtain the density quantification value multiplied by a weight of 0.3. According to the terrain slope data, the terrain quantitative value is obtained by assigning 20 points to the slope in the range of 0-5 degrees, 40 points to the slope in the range of 5-15 degrees, 60 points to the slope in the range of 15-25 degrees, 80 points to the slope in the range of 25-35 degrees, and 100 points to the slope greater than 35 degrees, and the terrain score is calculated by multiplying the weight coefficient 0.1. The membership of the quantitative value of each indicator is calculated by the membership function. The membership value is equal to the square of the score minus the minimum value divided by the square of the maximum value minus the minimum value using the quadratic curve function calculation method. The comprehensive score is obtained by weighted summation of the load score, density score, thickness score, and terrain score. The fire risk level distribution map is obtained based on the comprehensive score less than 30 points, which is judged as low fire risk, 30-50 points as medium-low fire risk, 50-70 points as medium fire risk, 70-90 points as medium-high fire risk, and more than 90 points as high fire risk. In the process of quantitative assessment of fire risk of litter layer, the area is divided into 30m×30m grid units through spatial rasterization processing, and the litter load value in each grid is quantitatively scored. When the load is 0.3 kg per square meter, it is assigned 20 points and multiplied by the weight 0.4 to get 8 points. When the load is 1.2 kg per square meter, it is assigned 40 points to get 16 points. When the load is 1.8 kg per square meter, it is assigned 60 points to get 24 points. When the load is 2.3 kg per square meter, it is assigned 80 points to get 32 ​​points. When the load is 2.8 kg per square meter, it is assigned 100 points to get 40 points.In the thickness and density scoring, the thickness of 1.5 cm is assigned 20 points multiplied by the weight of 0.2 to get 4 points, the thickness of 4 cm is assigned 40 points to get 8 points, the thickness of 8 cm is assigned 60 points to get 12 points, the thickness of 12 cm is assigned 80 points to get 16 points, the thickness of 18 cm is assigned 100 points to get 20 points, the density of 0.015 grams per cubic centimeter is assigned 20 points multiplied by the weight of 0.3 to get 6 points, the density of 0.035 grams per cubic centimeter is assigned 40 points to get 12 points, the density of 0.075 grams per cubic centimeter is assigned 60 points to get 18 points, the density of 0.125 grams per cubic centimeter is assigned 80 points to get 24 points, and the density of 0.18 grams per cubic centimeter is assigned 100 points to get 30 points. In the terrain scoring, a slope of 3 degrees is assigned 20 points multiplied by a weight of 0.1 to get 2 points, a slope of 12 degrees is assigned 40 points to get 4 points, a slope of 22 degrees is assigned 60 points to get 6 points, a slope of 32 degrees is assigned 80 points to get 8 points, and a slope of 38 degrees is assigned 100 points to get 10 points. The membership value is calculated by the quadratic curve function. When the score is 20 points, the membership is 0.04, 40 points, 0.16, 60 points, 0.36, 80 points, 0.64, and 100 points, the membership is 1.00. The weighted sum of the scores of the four indicators shows that the low fire risk area in the study area accounts for 15%, the medium-low fire risk area accounts for 25%, the medium fire risk area accounts for 35%, the medium-high fire risk area accounts for 18%, and the high fire risk area accounts for 7%.

[0024] Step S105, based on the classification of fire risk levels of the litter layer, analyze the spatial distribution characteristics of the fire risk of the litter layer, identify high-incidence areas of fire risk, and use spatial autocorrelation analysis methods to represent the spatial aggregation and correlation of the fire risk of the litter layer.

[0025] Obtain grid unit fire risk level data, calculate the difference value of fire risk level between adjacent units according to the fire risk level data, and use the distance decay function to construct a spatial weight matrix, wherein the distance decay function is w=1 / d², w is the weight between spatial units, and d is the distance between two spatial units; for the spatial weight matrix data, use the G statistic to calculate the local spatial autocorrelation index, identify the spatial clustering unit through the spatial autocorrelation index, and obtain the hot spot area distribution map from the Monte Carlo random permutation; perform distance segmentation according to the hot spot area distribution map, use the spatial autocorrelation index to calculate the correlation value of each distance segment, judge the significance of the correlation value through the Z test, and obtain the fire risk spatial correlation effect distance from the significance test result; for the spatial clustering unit, standardize the litter load, thickness, density, and terrain attributes, and use variance decomposition to calculate the contribution rate of standardized attributes to the spatial differentiation of fire risk; sort according to the contribution rate value, obtain the dominant factor of spatial differentiation from the sorting result, and judge the cause of spatial pattern through the dominant factor.

[0026] Specifically, the difference between the fire risk level of each grid unit and the adjacent unit is calculated based on the raster data. The spatial weight matrix is ​​constructed by setting the distance threshold of 300 meters. The distance decay function w=1 / d² is used to calculate the weight w between spatial units, where d is the distance between two units. The spatial correlation intensity value of the fire risk level is obtained from the weight matrix data. For the distribution map of fire risk levels in the region, the G statistic is used to calculate the local spatial autocorrelation index. The spatial clustering characteristics are identified by setting the search radius to 500 meters. The hot spot area distribution map under the significance level of 0.05 is obtained from 1000 Monte Carlo random permutations. The region is segmented according to the distance interval of 100 meters. The spatial autocorrelation index of each distance in the range of 0-1000 meters is calculated. The Z test is used to determine the significance of the association of each distance segment. The fire risk spatial correlation effect distance is obtained from the significance test results. The four attribute data of litter load, thickness, density, and terrain of the spatial unit are standardized. The contribution rate of the four attributes to the spatial differentiation of fire risk is calculated by variance decomposition. The dominant factor of spatial differentiation is obtained from the contribution rate ranking to determine the cause of the spatial pattern. In the process of analyzing the spatial characteristics of fire risk, the area was divided into 30m×30m grid cells. For each cell, the difference in fire risk level with the adjacent cells within 300m was calculated. When the distance between the two cells was 30m, the weight was 1.00, when the distance was 60m, the weight was 0.25, and when the distance was 90m, the weight was 0.11. The spatial correlation strength was obtained by weighted summation. When the correlation strength was greater than 0.8, it was highly correlated, when the correlation strength was between 0.5-0.8, it was moderately correlated, and when the correlation strength was less than 0.5, it was lowly correlated. In the local spatial autocorrelation analysis, a search radius of 500m was set to calculate the G statistic, and the significance test value was obtained by 1000 Monte Carlo random permutations. When the Z value was greater than 1.96, it was determined to be a significant cluster. Among them, the high fire risk area showed three significant hot spots, which were located in the northeast of the study area with an area of ​​5.2 square kilometers, the west with an area of ​​3.8 square kilometers, and the south with an area of ​​4.5 square kilometers. In the distance segmentation analysis, the spatial autocorrelation index was calculated every 100 meters. The autocorrelation index of the 0-100 meter segment was 0.82, 100-200 meter segment was 0.65, 200-300 meter segment was 0.48, 300-400 meter segment was 0.35, 400-500 meter segment was 0.22, and the autocorrelation index above 500 meters was less than 0.2. The Z test determined that 300 meters was the critical distance for significant spatial association. In the analysis of spatial differentiation driving forces, the contribution rates of the four attributes after standardized processing were 38% for litter load, 25% for thickness, 22% for density, and 15% for terrain, indicating that load distribution is the dominant factor in the formation of the spatial pattern of fire risk.

[0027] Step S106, analyzing the dynamic change characteristics of the fire risk of the litter layer through the fire risk assessment results of the litter layer, using the time series analysis method to summarize the seasonal change law and interannual change trend of the fire risk of the litter layer, and identifying the high-incidence period of fire risk.

[0028] According to the fire risk assessment data, trend item data, seasonal item data and random item data are decomposed by an additive model. The additive model uses an exponential smoothing coefficient to process random fluctuations, and obtains a long-term trend curve from a cubic polynomial fitting. Based on the trend item data, a sliding window is used to extract the inter-annual variation trend, and the slope value of the trend item data is calculated by linear regression. If the slope value is greater than a preset threshold, it is determined to be a stage of rapid increase in the fire risk level. Based on the seasonal item data, a spectrum analysis is performed, the spectrum data is smoothed by a Hanning window function, and the cycle length is determined from the peak position of the spectrum data. Based on the random item data, a three-level decomposition is performed using a wavelet basis function, and a fire risk mutation point is detected by setting a mutation threshold. If the amplitude of the mutation point exceeds the standard deviation threshold and the duration exceeds the duration threshold, it is determined to be a period of high fire risk. The month when fire risk is prone to occur is determined from the distribution law of the period of high fire risk, and the distribution law is obtained based on the time distribution characteristics of the mutation point.

[0029] Specifically, a time series data set is constructed based on the monthly fire risk assessment data. The time series is decomposed by the additive model. The seasonal cycle is set to 12 months. The exponential smoothing coefficient is 0.3 to deal with random fluctuations. The long-term trend curve is obtained from the cubic polynomial fitting. For the trend item data obtained by decomposition, a 12-month sliding window with a window sliding step of 1 month is used to extract the interannual variation trend. The slope change is calculated by linear regression. The monthly slope change is greater than 0.05, which is judged as the rapid increase stage of the fire risk level. The interannual variation law of fire risk is obtained from the slope change. Spectral analysis is performed based on the seasonal item data. The sampling frequency is set to 12 times per year and the spectrum window length is 36 months. The Hanning window function is used to smooth the spectrum data. The cycle length at the significance level of 0.05 is judged from the spectrum peak position. For the random item data, the db4 wavelet basis function is selected for three-level decomposition. The fire risk mutation point is detected by setting the mutation threshold to twice the standard deviation. When the mutation amplitude exceeds the threshold and the duration exceeds 2 months, it is judged as a high-incidence period of fire risk. The fire risk prone month is obtained from the distribution of the high-incidence period. In the analysis of the dynamic characteristics of fire risk changes, a long time series was constructed based on the monthly fire risk assessment data from 2020 to 2023. The data was divided into trend items, seasonal items and random items through additive model decomposition. The exponential smoothing coefficient was set to 0.3 to smooth the random fluctuations. The trend curve obtained by cubic polynomial fitting showed that the fire risk level showed a fluctuating upward trend, with an average of 2.5 in 2020 and rising to 3.2 in 2023. For the trend item, a 12-month sliding window and a step length of 1 month were used for trend extraction. The slope was calculated by linear regression and it was found that the monthly change rate from March to September 2021 exceeded 0.05, which showed a rapid increase in fire risk. In the periodic analysis, the sampling frequency was set to 12 times a year, and the window length of 36 months was selected for spectrum analysis. After smoothing with the Hanning window, two main cycles of 12 months and 6 months were detected at the significance level of 0.05. The 12-month cycle corresponds to the annual cycle change with an amplitude of 0.8 levels, and the 6-month cycle corresponds to the seasonal change with an amplitude of 0.3 levels. In the random fluctuation analysis, the db4 wavelet basis function was used for three-level decomposition, and twice the standard deviation was set as the mutation threshold. It was identified that the high-incidence periods lasting more than 2 months mainly occurred in August-October 2020, July-September 2021, September-November 2022, and August-October 2023, indicating that the high-incidence period of fire risk is mainly concentrated in the July-November period, with an average duration of 3 months.

[0030] Step S107, using the geographic detector method, quantitatively analyze the impact of terrain, weather, and vegetation on the fire risk of the litter layer, identify the dominant key factors, and explore the key factors affecting the fire risk of the litter layer.

[0031] Slope and aspect data are extracted according to digital elevation data, and the terrain differentiation factor value corresponding to the slope and aspect data is calculated by a geographic detector; the slope interval is divided according to the terrain differentiation factor value, and the meteorological station monitoring data is processed by the Kriging spatial interpolation method to obtain the continuous distribution result of the monitoring data; the temperature interval and the relative humidity interval are divided according to the continuous distribution result, and the normalized vegetation index data is classified by the natural breakpoint method to obtain the vegetation coverage classification result; the detection value corresponding to the terrain differentiation factor value, the continuous distribution result and the vegetation coverage classification result is calculated by a geographic detector; if the meteorological factor detection value is greater than the terrain factor detection value and greater than the vegetation factor detection value, the meteorological condition is determined to be the dominant factor of fire risk.

[0032] Specifically, the slope and aspect data were extracted from the 30-meter resolution digital elevation data, and the terrain differentiation factor value within the 1 square kilometer statistical unit was calculated through the geographic detector. The influence of terrain on fire risk was obtained from the five intervals of slope 0-5 degrees, 5-15 degrees, 15-25 degrees, 25-35 degrees, and above 35 degrees, and the detection value z statistic was calculated at a confidence level of 0.05. Kriging spatial interpolation was performed on the hourly monitoring data of the meteorological station, and continuous distribution data was obtained by setting the search radius of 500 meters, exponential variation function, and the minimum number of points to 12. The influence of meteorological factors on fire risk was calculated from the temperature classification intervals of 20-25℃, 25-30℃, 30-35℃, and 35-40℃, and the relative humidity classification intervals of 0-20%, 20-40%, 40-60%, and 60-80%. According to the normalized vegetation index data, the natural breakpoint method with a maximum number of iterations of 100, a classification fitness threshold of 0.85, an intra-class variance weight of 0.6, and an inter-class variance weight of 0.4 was used to classify vegetation coverage, and the influence of vegetation factors on fire risk was obtained from five intervals of vegetation coverage: 0-0.2, 0.2-0.4, 0.4-0.6, 0.6-0.8, and 0.8-1.0. The explanatory power of the three types of factors was calculated by the geographic detector under the conditions of a confidence level of 0.05, an interactive detection threshold of 0.1, an ecological detection z value of 1.96, and a risk detection f value of 0.05. The influence of terrain was obtained from the detection value of 0.45, the influence of meteorology was obtained from the detection value of 0.62, and the influence of vegetation was obtained from the detection value of 0.38. It was judged that meteorological conditions were the dominant factor of fire risk. In the quantitative analysis of geographic detectors, 30-meter resolution digital elevation was used to construct a 1 square kilometer statistical unit, and the terrain differentiation factor values ​​within the region were calculated. The area with a slope of 0-5 degrees accounted for 23%, with a detection value of 0.32, the area with a slope of 5-15 degrees accounted for 35%, with a detection value of 0.41, the area with a slope of 15-25 degrees accounted for 25%, with a detection value of 0.48, the area with a slope of 25-35 degrees accounted for 12%, with a detection value of 0.52, and the area with a slope of more than 35 degrees accounted for 5%, with a detection value of 0.56. The z statistic was 2.35 at a confidence level of 0.05. For the hourly monitoring data of 15 meteorological stations, a continuous distribution was obtained by Kriging interpolation with an exponential variation function, a search radius of 500 meters and a minimum number of 12 points. The detection value in the temperature area of ​​20-25℃ was 0.45, the detection value in the temperature area of ​​25-30℃ was 0.58, the detection value in the temperature area of ​​30-35℃ was 0.72, and the detection value in the temperature area of ​​35-40℃ was 0.83. The detection value of relative humidity in the area of ​​0-20% was 0.76, the detection value in the area of ​​20-40% was 0.65, the detection value in the area of ​​40-60% was 0.48, and the detection value in the area of ​​60-80% was 0.35.The vegetation index was classified by the natural breakpoint method with 100 iterations and a fitness threshold of 0.85. The detection value of the coverage area of ​​0-0.2 was 0.51, the detection value of the coverage area of ​​0.2-0.4 was 0.45, the detection value of the coverage area of ​​0.4-0.6 was 0.38, the detection value of the coverage area of ​​0.6-0.8 was 0.32, and the detection value of the coverage area of ​​0.8-1.0 was 0.25. The explanatory power of the three factors was calculated by the interactive detection threshold of 0.1 and the ecological detection z value of 1.96. The topographic detection value was 0.45, the meteorological detection value was 0.62, and the vegetation detection value was 0.38. Among them, the interactive detection value of meteorology and terrain was 0.71, the interactive detection value of meteorology and vegetation was 0.65, and the interactive detection value of terrain and vegetation was 0.52, indicating that the interaction between meteorological conditions and topography was the most significant.

[0033] Step S108, based on the fire risk assessment results of the litter layer, propose fire risk management measures for the litter layer, and formulate fire prevention measures for areas and periods with high fire risk. The fire prevention measures include strengthening litter cleaning, optimizing vegetation structure, and establishing fire isolation zones.

[0034] According to the terrain unit data, water system distribution data, road accessibility data and population density data, a fire risk level distribution map is obtained through overlay analysis, and areas with fire risk level values ​​greater than a preset threshold are extracted from the fire risk level distribution map to obtain a fire prevention management key area distribution map; for the fire prevention management key area distribution map, a robot cleaning device is used to perform cleaning operations along a preset path perpendicular to the slope direction, and a litter processing path map is obtained by setting a single cleaning area value, a cleaning thickness value and an operation interval value; according to the fire prevention management key area distribution map, a vegetation configuration database is constructed using a random forest algorithm, and the vegetation configuration database is extracted from the vegetation configuration database. Acquire the stand structure parameter values, and obtain the stand structure adjustment plan by setting the proportion of fire-resistant tree species, understory combustible coverage, canopy density and stand density. Use geographic information processing tools to construct fire isolation belts based on the boundaries of the distribution map of key fire prevention management areas, and obtain the fire isolation belt layout map by setting the isolation belt width value, the isolation belt direction value perpendicular to the dominant wind direction and the isolation belt spacing value. According to the fire isolation belt layout map, obtain road distribution information from existing road network data, establish the connectivity between the fire isolation belt and the existing road network by setting the fire road width value, and obtain a complete fire prevention management unit distribution map.

[0035] Specifically, fire prevention management units are established based on the distribution map of fire risk levels in high-incidence areas. By superimposing terrain units, water system distribution, road accessibility, and population density data, the distribution map of key fire prevention management areas is obtained from areas with fire risk levels greater than level 4, slopes greater than 25 degrees, population densities greater than 100 people per square kilometer, and road density less than 2 kilometers per square kilometer. For key fire prevention management areas, a robot cleaning device with an operating speed of 2 meters per second and a cleaning width of 1.5 meters is used. The litter treatment path map is obtained by setting the cleaning path perpendicular to the slope, a single cleaning area of ​​0.1 square kilometers, a cleaning thickness of 2 centimeters, and an operation interval of 7 days. According to the relationship between fire risk and vegetation structure, a vegetation configuration database with a proportion of 80% fire-resistant tree species, a coverage rate of 20% for understory combustibles, a canopy closure of 0.6, and a forest density of 0.8 plants per square meter is constructed through the random forest algorithm, and the forest structure adjustment plan is obtained from the optimization of vegetation parameters. For the boundaries of fire prevention management units, geographic information processing tools were used to construct fire isolation belts. The layout map of fire isolation belts was obtained by setting the width of the isolation belt to 20 meters, the isolation belt perpendicular to the dominant wind direction, the isolation belt spacing to 200 meters, the fire road width to 4 meters, and connecting with the existing road network. In the process of implementing fire prevention management in the study area, management unit division was carried out based on the fire risk level distribution map. When the fire risk level was level 4, the slope exceeded 25 degrees, the population density reached 150 people per square kilometer, and the road density was less than 1.5 kilometers per square kilometer, the area was designated as a key fire prevention area, of which the northeastern mountainous area covered an area of ​​5.2 square kilometers, the western mountainous area covered an area of ​​3.8 square kilometers, and the southern mountainous area covered an area of ​​4.5 square kilometers. For the cleaning of fallen leaves in key areas, crawler-type automatic cleaning robots are deployed, with an operating speed of 2 meters per second, the cleaning path is laid out along the contour line, the cleaning width is 1.5 meters, the single operation area is controlled at 0.1 square kilometers, the cleaning depth is 2 centimeters, and the operation is repeated every 7 days. The cleaned materials are concentrated at designated stacking points beside the fire prevention road. In the optimization of vegetation configuration, a fire risk prediction model is established through the random forest algorithm, and a forest stand configuration scheme is selected with a fire-resistant tree species accounting for 80%, the coverage rate of combustible materials under the forest controlled below 20%, the canopy density maintained at 0.6, and the spacing between rows and plants adjusted to 1.25 meters to form a fire prevention vegetation corridor. A fire isolation system is constructed at the boundary of the management unit, with the width of the isolation belt set to 20 meters, the strip direction perpendicular to the northeast dominant wind direction, the distance between adjacent isolation belts is 200 meters, and the width of the fire prevention road is 4 meters. It is connected with the existing highway to construct a ring-shaped fire prevention channel network to form a fire barrier.

[0036] Step S109, pre-establish a litter layer fire risk monitoring and early warning system, obtain litter condition information in real time, combine meteorological data, use support vector machine algorithm to dynamically warn of litter layer fire risk, and issue fire risk early warning information if there is a litter layer fire risk.

[0037] The temperature and humidity monitoring data are obtained through ridge line sensors, valley line sensors and slope sensors; the litter state parameters of the area are collected according to the temperature and humidity monitoring data, and the state parameters include moisture content data obtained by the biomass moisture sensor and load data obtained by the optical sensor; a support vector machine prediction model is established for the temperature and humidity monitoring data and litter state parameters, and the input features of the prediction model include temperature parameters, humidity parameters, wind speed parameters, moisture content parameters, load parameters, slope parameters, slope direction parameters, and altitude parameters; according to the radial basis function kernel function and A fire risk prediction model is obtained by penalty parameter training. If the verification accuracy of the fire risk prediction model reaches a preset threshold, a fire risk prediction result is obtained. A fire risk index is calculated based on the fire risk prediction result. The fire risk index is determined by a temperature weight parameter, a humidity weight parameter, a load weight parameter, and a wind speed weight parameter. If the fire risk index is in a first preset interval, it is determined to be a yellow warning level; if the fire risk index is in a second preset interval, it is determined to be an orange warning level; if the fire risk index is in a third preset interval, it is determined to be a red warning level; and if the fire risk index is in a fourth preset interval, it is determined to be a purple warning level.

[0038] Specifically, the sensor monitoring network is deployed according to the terrain units in the monitoring area. Sensors are set up at three types of landforms: ridge lines, valley lines, and slopes. Meteorological sensors with a temperature accuracy of 0.1°C, a humidity accuracy of 1%, and a sampling frequency of 5 minutes are used to obtain temperature and humidity monitoring values ​​from the ZigBee communication protocol. For litter sensor monitoring data, a biomass moisture sensor with a moisture measurement range of 0-100% and an optical sensor with a resolution of 640×480 are used to collect litter moisture content and load data. The litter state parameters are obtained through a data compression rate of 0.6, a storage period of 1 hour, and a sampling frequency of 5 minutes. The support vector machine prediction model is trained based on real-time monitoring data. Through 8 input features including temperature, humidity, wind speed, moisture content, load, slope, slope direction, and altitude, the kernel function radial basis function, penalty parameter 1.0, cross-validation ratio 0.2, number of iterations 1000, and validation accuracy 0.85 are set to obtain a fire risk prediction model. Fire risk warning rules are established based on the prediction results. The fire risk index is calculated by temperature weight 0.3, humidity weight 0.3, load weight 0.2, and wind speed weight 0.2. When the index is 0.8-0.85, it is judged as a yellow warning, 0.85-0.9 as an orange warning, 0.9-0.95 as a red warning, and 0.95-1.0 as a purple warning, the warning information is issued within 5 minutes of the response time. During the implementation of regional fire risk monitoring and early warning, 15 monitoring points are set up on the ridge line, 12 monitoring points are set up on the valley line, and 25 monitoring points are set up on the slope according to the terrain characteristics. The monitoring equipment includes meteorological sensors with a temperature accuracy of 0.1°C and a humidity accuracy of 1%. Data is transmitted every 5 minutes through the Zigbee protocol. The measured temperature data is in the range of 25-35°C and the relative humidity is in the range of 20-60%. In the monitoring of litter status, a biomass moisture sensor with a measurement range of 0-100% is used to detect moisture content, and an optical sensor with a resolution of 640×480 obtains load images. The monitoring data shows that the moisture content of litter is distributed in the range of 8-25%, and the load is distributed in the range of 0.5-2.5 kg per square meter. The data is stored with a compression rate of 0.6, and a data file is formed every hour. In the fire risk prediction modeling, eight features are selected, including temperature, humidity, wind speed, moisture content, load, slope, slope direction, and altitude. The prediction model is established through support vector machine. The kernel function uses radial basis function, the penalty parameter is set to 1.0, and the cross-validation ratio of 0.2 is used to train 1000 groups of sample data, and the verification accuracy reaches 0.85. In the judgment of early warning rules, the fire risk index is calculated based on the temperature weight of 0.3, humidity weight of 0.3, load weight of 0.2, and wind speed weight of 0.2, among which the yellow warning area accounts for 15%, the orange warning area accounts for 8%, the red warning area accounts for 5%, and the purple warning area accounts for 2%. The early warning information is released to the designated terminal through the monitoring network within 5 minutes.

[0039] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A control method for a forest fire early warning system, characterized in that: The method comprises: The spatial distribution data of litter in the area are obtained through manual inspection and reporting, drone monitoring and satellite remote sensing fire monitoring, including the coverage, thickness and density of litter. A litter spatial distribution database is established based on the obtained data to obtain the spatial distribution characteristics of litter. Combustion characteristic tests are carried out for different types of litter to determine the ignition point, calorific value and burning rate parameters of litter to obtain the combustion characteristic data of different types of litter. According to the spatial distribution characteristics and combustion characteristics of litter, a litter layer fire risk assessment index system is constructed, and the key factors affecting the fire risk, including litter load, litter thickness, litter density and terrain factors, are combined to determine the dominant key factors using the hierarchical analysis method. The constructed litter layer fire risk assessment index system is used to quantitatively assess the fire risk of the litter layer in the study area, and the fuzzy comprehensive evaluation method is used to quantify and score each assessment index to obtain the litter layer fire risk level distribution map. On the basis of the litter layer fire risk level division, the litter layer fire risk level is analyzed. The spatial distribution characteristics of the fire risk of the litter layer are used to identify the high-incidence areas of fire risk. The spatial autocorrelation analysis method is used to represent the spatial clustering and correlation of the fire risk of the litter layer. The dynamic change characteristics of the fire risk of the litter layer are analyzed through the fire risk assessment results of the litter layer. The seasonal change law and interannual change trend of the fire risk of the litter layer are summarized by the time series analysis method, and the high-incidence period of fire risk is identified. The geographic detector method is used to quantitatively analyze the influence of topography, meteorology, and vegetation on the fire risk of the litter layer, identify the dominant key factors, and explore the key factors affecting the fire risk of the litter layer. Based on the fire risk assessment results of the litter layer, fire risk management measures for the litter layer are proposed, and fire prevention measures are formulated for the high-incidence areas and high-incidence periods of fire risk. The fire prevention measures include strengthening litter cleaning, optimizing vegetation structure, and establishing fire isolation belts. A fire risk monitoring and early warning system for the litter layer is established in advance to obtain litter status information in real time. Combined with meteorological data, the support vector machine algorithm is used to dynamically warn the fire risk of the litter layer. If there is a fire risk in the litter layer, a fire risk warning information will be issued.

2. The method according to claim 1, characterized in that The spatial distribution data of litter in the area is obtained by manual inspection and reporting, drone monitoring and satellite remote sensing fire monitoring, including the coverage, thickness and density of litter, and a litter spatial distribution database is established based on the obtained data to obtain the litter spatial distribution characteristics, including: A multispectral camera is mounted on an unmanned aerial vehicle to obtain a surface image, spectral data of a visible light band and a near infrared band are extracted from the surface image, and a standardized vegetation index is calculated based on the spectral data; Establishing a linear regression correlation model based on the standardized vegetation index and the measured litter thickness value on the surface, and obtaining litter distribution data from the linear regression correlation model; For the surface reflectance data in the satellite remote sensing image, the spectral similarity value is calculated using the near-infrared band and the infrared band spectral curves. If the spectral similarity value is greater than a preset threshold, it is determined to be a litter covered area; The high-resolution data collected by the UAV is resampled by principal component transformation, and the spatial registration parameters are obtained through ground control points. The spatial correspondence between the UAV data and the satellite data is realized according to the spatial registration parameters; The vegetation index change value was calculated based on the time series remote sensing image sequence, and the seasonal change characteristic curve was extracted using the time series decomposition method. The litter accumulation period was determined based on the slope and inflection point position of the change characteristic curve.

3. The method according to claim 1, characterized in that The combustion characteristic test is carried out for different types of litter, and the ignition point, calorific value, and combustion rate parameters of the litter are measured to obtain the combustion characteristic data of different types of litter, including: The temperature data output by the litter surface temperature sensor and the internal temperature data output by the optical fiber temperature sensor are collected. If the temperature data rising rate exceeds the preset temperature threshold and the duration exceeds the preset time threshold, the litter ignition point temperature value is obtained; According to the heat flux data output by the heat flux sensor and the surface radiation intensity data output by the infrared spectrometer, when the heat flux data is higher than a preset heat flux threshold and the duration exceeds a preset duration, a heat value parameter is obtained; For an image sequence captured by an infrared thermal imager, several principal components of the image sequence are extracted to obtain heat source characteristic data, a combustion area boundary is divided according to the heat source characteristic data, and the combustion rate is calculated from the combustion area boundary; Collect oxygen concentration data, carbon monoxide concentration data and carbon dioxide concentration data output by the gas sensor, and establish a gas parameter data set based on the gas concentration data; The temperature data, the heat flux data and the gas parameter data set are subjected to multi-parameter correction by Kalman filtering, and combustion characteristic parameters are obtained from the corrected data.

4. The method according to claim 1, characterized in that According to the spatial distribution characteristics and combustion characteristics of litter, a litter layer fire risk assessment index system is constructed, and the key factors affecting fire risk, including litter load, litter thickness, litter density, and terrain factors, are combined. The hierarchical analysis method is used to determine the dominant key factors, including: Receiving multispectral remote sensing data of near infrared band and red light band, calculating a normalized vegetation index according to the multispectral remote sensing data, and establishing a regression equation through the normalized vegetation index and litter load; The regression equation is used to calculate the spatial distribution value of litter load, generate an elevation map, and obtain the slope classification weight value from the elevation map; Based on the litter thickness data of the region, the spherical semivariogram and Kriging interpolation were used to calculate the thickness spatial distribution value, where the spherical semivariogram is γ(h)=C0+C, where C0 is the nugget value, C is the base value, and h is the sampling interval; The unit area weight density value is calculated based on the thickness spatial distribution value, and the density value is obtained by a weight calculation formula; A judgment matrix is ​​constructed for the load value, density value, thickness value and terrain weight value. If the consistency ratio is less than a preset threshold, the weight calculation result is judged to be reasonable, and the key factor evaluation index is obtained from the judgment matrix.

5. The method according to claim 1, characterized in that The constructed litter layer fire risk assessment index system is used to quantitatively assess the litter layer fire risk in the study area, and the fuzzy comprehensive evaluation method is used to quantify and score each assessment index to obtain a litter layer fire risk level distribution map, including: Obtaining a load interval score according to the litter load value, wherein the load interval score is determined by the corresponding relationship of the litter load value within a preset interval, and multiplied by the load weight coefficient to obtain a load score; Obtaining a thickness interval score for the litter thickness value, wherein the thickness interval score is determined by the corresponding relationship of the litter thickness value within a preset interval and multiplied by a thickness weight coefficient to obtain a thickness score; At the same time, a density interval score is obtained for the litter density value, wherein the density interval score is determined by the corresponding relationship of the litter density value within a preset interval, and is multiplied by the density weight coefficient to obtain a density score; Obtaining a slope interval score according to the terrain slope value, wherein the slope interval score is determined by a corresponding relationship between the terrain slope values ​​within a preset interval and multiplied by a slope weight coefficient to obtain a terrain score; A quadratic curve function is used to calculate the membership values ​​of the load score, the density score, the thickness score and the terrain score, wherein the membership value is obtained by dividing the square of the interval score minus the minimum value by the square of the maximum value minus the minimum value; Obtaining a comprehensive score by weighted sum of the load score, the density score, the thickness score and the terrain score, and if the comprehensive score is less than or equal to a first threshold, determining the fire risk level to be low; If the comprehensive score is greater than the first threshold and less than or equal to the second threshold, it is determined to be a medium-low fire risk level; If the comprehensive score is greater than the second threshold and less than or equal to the third threshold, it is determined to be a medium fire risk level; If the comprehensive score is greater than the third threshold and less than or equal to the fourth threshold, it is determined to be a medium-high fire risk level; If the comprehensive score is greater than a fourth threshold, it is determined to be a high fire risk level.

6. The method according to claim 1, characterized in that On the basis of the classification of fire risk levels of litter layer, the spatial distribution characteristics of fire risk of litter layer are analyzed to identify the high-incidence areas of fire risk. The spatial autocorrelation analysis method is used to express the spatial aggregation and correlation of fire risk of litter layer, including: Obtain grid unit fire risk level data, calculate the fire risk level difference between adjacent units based on the fire risk level data, and construct a spatial weight matrix using a distance decay function, where the distance decay function is w=1 / d², w is the weight between spatial units, and d is the distance between two spatial units; For the spatial weight matrix, the local spatial autocorrelation index is calculated using the G statistic, the spatial clustering unit is identified through the spatial autocorrelation index, and the hot spot area distribution map is obtained from the Monte Carlo random permutation; Distance segmentation is performed according to the hot spot area distribution map, the spatial autocorrelation index is used to calculate the correlation value of each distance segment, the significance of the correlation value is judged by Z test, and the fire risk spatial correlation effect distance is obtained from the significance test result; For the spatial clustering units, the litter load, thickness, density and terrain attributes were standardized, and the contribution rate of the standardized attributes to the spatial differentiation of fire risk was calculated using variance decomposition. The contribution rate is used to sort the data, and the dominant factors of spatial differentiation are obtained from the sorting results. The causes of spatial patterns are determined by the dominant factors.

7. The method according to claim 1, characterized in that The fire risk assessment results of the litter layer are used to analyze the dynamic change characteristics of the fire risk of the litter layer. The seasonal change law and interannual change trend of the fire risk of the litter layer are summarized by using the time series analysis method, and the high-incidence period of fire risk is identified, including: According to the fire risk assessment data, trend item data, seasonal item data and random item data are obtained by decomposing them through an additive model, wherein the additive model uses an exponential smoothing coefficient to process random fluctuations and obtains a long-term trend curve from a cubic polynomial fitting; Based on the trend item data, a sliding window is used to extract the inter-annual variation trend, and a slope value of the trend item data is calculated by linear regression. If the slope value is greater than a preset threshold, it is determined to be a stage of rapid increase in fire risk level; Performing spectrum analysis based on the seasonal item data, smoothing the spectrum data using a Hanning window function, and determining the cycle length from the peak position of the spectrum data; Based on the random item data, a wavelet basis function is used to perform a three-level decomposition, and a fire risk mutation point is detected by setting a mutation threshold. If the amplitude of the mutation point exceeds the standard deviation threshold and the duration exceeds the duration threshold, it is determined to be a period of high fire risk; The months in which fire risks are prone to occur are determined from the distribution pattern of the high-incidence periods of fire risks, wherein the distribution pattern is obtained based on the time distribution characteristics of the mutation points.

8. The method according to claim 1, characterized in that The geographic detector method is used to quantitatively analyze the influence of topography, weather, and vegetation on the fire risk of litter layer, identify the dominant key factors, and explore the key factors affecting the fire risk of litter layer, including: Extracting slope and aspect data according to digital elevation data, and calculating terrain differentiation factor values ​​corresponding to the slope and aspect data through a geographic detector; The slope intervals are divided according to the terrain differentiation factor values, and the meteorological station monitoring data are processed using the Kriging spatial interpolation method to obtain the continuous distribution results of the monitoring data; Dividing the temperature interval and the relative humidity interval according to the continuous distribution result, and grading the normalized vegetation index data by the natural breakpoint method to obtain the vegetation coverage grading result; Calculating the detection values ​​corresponding to the terrain differentiation factor value, the continuous distribution result and the vegetation coverage classification result by means of a geographic detector; If the meteorological factor detection value is greater than the terrain factor detection value and greater than the vegetation factor detection value, the meteorological condition is judged to be the dominant factor of fire risk.

9. The method according to claim 1, characterized in that: Based on the fire risk assessment results of the litter layer, fire risk management measures for the litter layer are proposed, and fire prevention measures are formulated for areas and periods with high fire risk. The fire prevention measures include strengthening litter cleaning, optimizing vegetation structure, and establishing fire isolation zones, including: According to the terrain unit data, water system distribution data, road accessibility data and population density data, a fire risk level distribution map is obtained through overlay analysis, and areas with fire risk level values ​​greater than a preset threshold are extracted from the fire risk level distribution map to obtain a distribution map of key fire prevention management areas; According to the distribution map of key fire prevention management areas, a robot cleaning device is used to perform cleaning operations along a preset path perpendicular to the slope direction, and a litter treatment path map is obtained by setting a single cleaning area value, a cleaning thickness value, and an operation interval value; According to the distribution map of the key fire prevention management areas, a vegetation configuration database is constructed using a random forest algorithm, forest stand structure parameter values ​​are obtained from the vegetation configuration database, and a forest stand structure adjustment plan is obtained by setting the fire-resistant tree species ratio value, the understory combustible coverage value, the canopy density value, and the forest stand density value; According to the boundaries of the distribution map of key fire prevention management areas, geographic information processing tools are used to construct fire isolation belts, and the layout map of fire isolation belts is obtained by setting the isolation belt width value, the isolation belt direction value perpendicular to the dominant wind direction, and the isolation belt spacing value; According to the fire isolation zone layout map, road distribution information is obtained from existing road network data, and the connectivity between the fire isolation zone and the existing road network is established by setting the fire road width value to obtain a complete fire management unit distribution map.

10. The method according to claim 1, characterized in that The pre-established litter layer fire risk monitoring and early warning system obtains litter condition information in real time, combines meteorological data, and uses a support vector machine algorithm to dynamically warn of litter layer fire risk. If there is a litter layer fire risk, fire risk early warning information is issued, including: Acquire temperature and humidity monitoring data through ridge line sensors, valley line sensors and slope sensors; Collecting regional litter state parameters according to the temperature and humidity monitoring data, wherein the state parameters include moisture content data obtained by a biomass moisture sensor and load data obtained by an optical sensor; A support vector machine prediction model is established for the temperature and humidity monitoring data and litter state parameters, wherein the input features of the prediction model include temperature parameters, humidity parameters, wind speed parameters, moisture content parameters, load parameters, slope parameters, slope aspect parameters and altitude parameters; A fire risk prediction model is obtained according to the radial basis function kernel function and penalty parameter training of the support vector machine prediction model, and a fire risk prediction result is obtained if the verification accuracy of the fire risk prediction model reaches a preset threshold; A fire risk index is calculated based on the fire risk prediction result. The fire risk index is determined by a temperature weight parameter, a humidity weight parameter, a load weight parameter and a wind speed weight parameter. If the fire risk index is in a first preset interval, it is determined to be a yellow warning level. If the fire risk index is in a second preset interval, it is determined to be an orange warning level. If the fire risk index is in a third preset interval, it is determined to be a red warning level. If the fire risk index is in a fourth preset interval, it is determined to be a purple warning level.

Images