Forest fire local wind field downscaling prediction method and system considering fire-atmosphere interaction

Abstract

The invention provides a forest fire local wind field downscaling prediction method and a forest fire local wind field downscaling prediction system considering fire-atmosphere interaction. The method comprises the following steps: acquiring meteorological attribute time sequence data of different orientations of a fire scene, spatial positions of meteorological stations and fire spreading process infrared image time sequence data of remote sensing observation; approximately determining a fire head direction, and further determining directions of a fire tail and two side wings; determining meteorological data close to a fire head, meteorological data close to a fire tail and meteorological data close to two side wings; extracting time sequence data of a fire spreading situation descriptive index through infrared image data of a fire spreading process; establishing a deep learning model, and training a four-azimuth wind speed and direction prediction model of the forest fire local; and through an interpolation method, outputting data based on the fire scene four-direction wind speed and direction prediction model, estimating the wind speed and direction of each position in the fire scene, and realizing wind field downscaling.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent prevention and control of forest fires, and specifically is a method and system for downscaling prediction of local wind fields of forest fires taking into account fire-atmosphere interaction. Background Art

[0002] In recent years, global warming has intensified, extreme weather has occurred frequently, and forest fires have become more and more frequent. Therefore, in order to reduce the losses caused by wildfires to humans, it is urgent to develop intelligent forest fire prevention and control tools and systems. The role of wind in forest fires is very important because it can affect the origin, spread and control of fires, making fires more destructive and dangerous. Therefore, in the process of fire prevention and response, understanding and monitoring the speed and direction of wind are crucial for effective management and control of fires. However, there is a complex interaction between the spread of fire and the wind field during the development of fires. For example, large-scale fires will produce strong thermal air currents, which rise to form convection and cause drastic changes in wind speed and direction; strong winds will cause fresh air to flow into the fire area, increase oxygen supply, make the flames easier to burn, and make the fire more fierce. Therefore, to estimate the local wind field of the fire site, it is necessary to consider the disturbance of the wind field by the fire spread situation and conduct in-depth research on the fire-atmosphere interaction phenomenon.

[0003] At present, wind speed and wind direction prediction has attracted widespread attention in many fields. Among them, prediction methods based on artificial intelligence have been increasingly valued and become one of the most effective prediction methods. The main idea is to mine the time-varying trend of wind speed or wind direction from historical wind speed data to predict future wind speed. However, due to the disturbance of fire-atmosphere interaction, these methods are often not effectively used for downscaling prediction of local wind fields in fire areas. The key is that they cannot make full use of fire spread situation data to support local wind field modeling in fire areas. Summary of the invention

[0004] In view of the difficult problems mentioned in the above background technology, the purpose of the present invention is to provide a method and system for downscaling prediction of local wind field in a fire scene by integrating fire spread situation data and historical meteorological data as input. This method expands the dimension of input data of wind speed and wind direction prediction model; establishes a multivariate time series deep learning model; predicts wind speed and wind direction values ​​at key locations of the fire scene based on historical meteorological data and descriptive indicator data of fire spread situation; further, interpolates wind vectors at various locations inside the fire scene based on wind speed and wind direction values ​​in four directions. It provides data support for forest fire fighting decision-making.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A downscaling prediction method for local wind field of forest fire considering fire-atmosphere interaction comprises the following steps:

[0007] Step 1: Obtain the time series data of meteorological attributes at different locations of the fire scene, the spatial location of each meteorological station, and the time series data of infrared images of the fire spread process observed by remote sensing.

[0008] Step 2: According to the direction of change of the temperature center of gravity during the fire spread process, the direction of the fire head is approximately determined, and then the direction of the fire tail and the two side wings is determined.

[0009] Step 3: Calculate the distance between the position of each meteorological station in step 1 and the straight lines in the four directions in step 2, and select the monitoring data of the meteorological station with the shortest distance to each straight line as the meteorological data near the fire head, the meteorological data near the fire tail, and the meteorological data near the two wings.

[0010] Step 4: Extract the time series data of descriptive indicators of fire spread situation through the infrared image data of the fire spread process.

[0011] Step 5: Based on the time series data of meteorological attributes of the four directions of the fire scene obtained in step 3 and the time series data of descriptive indicators of the fire spread situation obtained in step 4, the final data set is established.

[0012] Step 6: Establish a deep learning model and train a four-directional wind speed and direction prediction model for local forest fires based on the time series data set described in step 5.

[0013] Step 7: Using the Kriging interpolation method, based on the output data of the four-direction wind speed and direction prediction model of the fire scene, estimate the wind speed and direction at each location inside the fire scene, and finally achieve wind field downscaling.

[0014] Furthermore, the meteorological attributes described in step 1 include wind speed, wind direction, temperature, humidity, and air pressure; the meteorological stations described in step 1 should be evenly distributed around the fire scene, and the number of meteorological stations N>=4; the infrared image pixel value described in step 1 is equal to the temperature of the pixel point.

[0015] Furthermore, in step 2, the direction of the fire head is approximately determined according to the change direction of the temperature center of gravity during the fire spread process, and the specific process of determining the direction of the fire tail and the two side wings is as follows:

[0016] Step 2.1: Binarize the infrared image using an adaptive temperature threshold, mark the area consisting of pixels with a temperature higher than the threshold as Burned, and mark the area consisting of pixels with a temperature lower than the threshold as Unburned.

[0017] Step 2.2: Take the temperature values ​​of all pixels in the Burned area as the mass of the point and calculate the center of gravity (x, y) of the area.

[0018] Step 2.3: Assume that the center of gravity of the fire scene infrared image at time t is (xt ,y t ), the center of gravity of the fire scene infrared image at time t+1 is (x t+1 ,y t+1 ), then the direction of the fire head at time t is represented by the vector (x t+1 -x t ,y t+1 -y t ) direction; the direction of the fire tail is in the opposite direction, and the directions of the two side wings form angles of 90° and 270° with the direction of the fire head respectively.

[0019] Furthermore, the fire spread situation descriptive indicators described in step 4 include: the perimeter L of the burned area, the surface area S, the change in the horizontal coordinate of the center of gravity Δx, the change in the vertical coordinate Δy, the perimeter elongation r L , area expansion rate r S . r L , and r S The calculation formula is as follows:

[0020]

[0021] Where, L t+1 is the perimeter of the Burned area at time t+1; L t is the perimeter of the Burned area at time t;

[0022] S t+1 is the area of ​​the Burned region at time t+1; S t is the area of ​​the Burned region at time t.

[0023] Furthermore, the final data set described in step 5 is an M*N-order matrix, where M is the sum of the number of meteorological attribute variables and the descriptive indicators of the fire spread situation, and N is the number of samples, that is, the time dimension.

[0024] Furthermore, the deep learning model described in step 6 is inputted with the 4-directional meteorological attribute data of the fire scene and the descriptive indicator data of the fire spread situation at the previous n moments, and the output is the 4-directional wind speed and direction values ​​at the n+1th moment.

[0025] Furthermore, the deep learning model described in step 6 includes but is not limited to a CNN-LSTM model.

[0026] A downscaling prediction system for local wind fields of forest fires considering fire-atmosphere interaction, including:

[0027] The acquisition module is used to obtain the time series data of meteorological attributes at different locations of the fire scene, the spatial location data of each meteorological station, and the time series data of infrared images of the fire spread process.

[0028] The calculation module is used to calculate the descriptive index of the fire spread situation according to the infrared image data of the fire spread process.

[0029] The key location meteorological data selection module is used to determine the spread direction of the fire head, fire tail and two wings according to the change of the temperature center of gravity during the fire spread process, and select the meteorological data near the fire head, fire tail and side wings according to the distance from the meteorological station to the straight line of each spread direction.

[0030] The prediction module is used to make predictions based on the four-directional meteorological data and the descriptive indicator data of the fire spread situation at the previous n moments, and obtain the prediction results of the four-directional wind speed and wind direction values ​​at the n+1th moment.

[0031] Interpolation module. It is used to interpolate the wind speed and wind direction values ​​of the four directions of the fire scene at the current moment, obtain the wind speed and wind direction values ​​of each position of the fire scene, and thus construct the local wind field of the fire scene.

[0032] Compared with the prior art methods, the present invention has the following beneficial effects:

[0033] The present invention provides a downscaling prediction method for the local wind field of forest fires considering the interaction between fire and atmosphere. The method is based on the multi-directional meteorological monitoring data of the fire scene and the infrared remote sensing image data of the fire spread process. The fire head, fire tail and two wings are determined according to the change of the temperature center of gravity of the fire scene in the infrared image, and then the data recorded by four meteorological stations near the fire head, fire tail and two wings are selected as the meteorological attribute input of the wind speed and wind direction prediction model. In addition, according to the changes in the shape and position of the fire scene in the infrared image, the indicators used to describe the fire spread situation, such as the perimeter of the fire scene and the area of ​​the fire scene, are extracted as another part of the input of the model. By expanding the dimension of the input data, in the prediction of the local wind speed and wind direction of the fire scene, not only the autocorrelation of the sequence is paid attention to, but also the interaction between fire and atmosphere is considered. The influence of the fire scene state on the wind vector is explored, and the descriptive indicators of the fire spread situation are extracted to support the wind vector modeling, which provides a new solution for the prediction of the local wind vector of the fire scene. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A flowchart of a method for downscaling prediction of local wind fields of forest fires taking into account fire-atmosphere interaction according to the present invention;

[0035] Figure 2 This is a schematic diagram of the calculation principle of the fire head direction of a fire scene at consecutive moments in an embodiment of the present invention;

[0036] Figure 3 This is a structural diagram of a deep learning model in an embodiment of the present invention;

[0037] Figure 4 This is a diagram of the interpolation result of the local wind speed at the fire scene in an embodiment of the present invention;

[0038] Figure 5 This is a diagram of the interpolation result of the local wind direction at the fire scene in an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0040] Combination Figure 1 The first object of the present invention is to provide a downscaling prediction method for local wind field of forest fire considering fire-atmosphere interaction, which specifically comprises the following steps:

[0041] Step 1: Obtain the time series data of meteorological attributes at different locations of the fire scene, including wind speed, wind direction, temperature, humidity, and air pressure; the location of each meteorological station, and the time series data of infrared images of the fire spread process observed by remote sensing, and establish the initial data set M. Its data structure is shown in Table 1:

[0042] Table 1

[0043]

[0044] Step 2: Based on the temperature corresponding to each pixel value of the fire scene infrared image, calculate the center of gravity coordinates of the burning infrared image at the current moment, combined with Figure 2 The directions of the fire head, fire tail and two wings are determined according to the vector formed by the temperature center of gravity of the remote sensing image at the current moment and the temperature center of gravity of the remote sensing image at the next moment.

[0045] 2.1) Binarize the infrared image using an adaptive temperature threshold, mark the area consisting of pixels with a temperature higher than the threshold as Burned, and mark the area consisting of pixels with a temperature lower than the threshold as Unburned;

[0046] 2.2) Take the temperature values ​​of all pixels in the Burned area as the mass of the point and calculate the center of gravity (x, y) of the area;

[0047] 2.3) Let the center of gravity of the infrared image of the fire scene at time t be (x t ,y t ), the center of gravity of the fire scene infrared image at time t+1 is (x t+1 ,y t+1 ), then the direction of the fire head at time t is represented by the vector (x t+1 -x t ,y t+1 -y t ) direction; the direction of the fire tail is in the opposite direction, and the directions of the two side wings form angles of 90° and 270° with the direction of the fire head respectively.

[0048] Step 3: Calculate the distance between the current position of each meteorological station and the straight lines where the fire head, fire tail and two wings are located, and select the meteorological data collected by the meteorological station closest to each straight line as the meteorological attributes near the fire head, meteorological attributes near the fire tail, meteorological attributes near the left wing and meteorological attributes near the right wing.

[0049] Step 4: Using image processing technology, the following fire spread situation descriptive indicators corresponding to each moment are calculated based on the infrared remote sensing image of the fire spread process: Burned area perimeter L, surface area S, center of gravity horizontal coordinate change Δx, vertical coordinate change Δy, perimeter elongation r L , area expansion rate r S . r L , and r S The calculation formula is as follows:

[0050]

[0051] Where, L t+1 is the perimeter of the Burned area at time t+1; L t is the perimeter of the Burned area at time t;

[0052] S t+1 is the area of ​​the Burned region at time t+1; S t is the area of ​​the Burned region at time t.

[0053] Step 5: The final data set is established based on the meteorological attribute data of the key positions (fire head, fire tail, left wing, and right wing) selected in step 3 and the descriptive index values ​​of the fire spread situation at each moment during the spread process. The data structure of the final data set is shown in Table 2.

[0054] Table 2

[0055]

[0056] Step 6: Establish a deep learning model. The input of the model is the 4-direction meteorological attribute data of the fire scene and the descriptive indicator data of the fire spread situation at the previous n moments, and the output is the 4-direction wind speed and direction values ​​at the n+1th moment. Based on the time series data set described in step 5, set the training set and test set in a certain ratio to carry out training. After completion, it can be realized to predict the 4-direction wind speed and wind direction values ​​of the fire scene for a certain time length in the future. Combined with Figure 3 In this embodiment of the present invention, a CNN-LSTM model is selected as an example, but the deep learning model described in step 6 includes but is not limited to a CNN-LSTM model.

[0057] Step 7: By using the interpolation method, based on the output data of the wind speed and direction prediction model of the four directions of the fire scene, the wind speed and direction of each position inside the fire scene are estimated, and finally the wind field is downscaled. In the embodiment of the present invention, the Kriging interpolation method is selected, but the interpolation method described in step 7 includes but is not limited to the Kriging interpolation method. The wind speed interpolation result of the target fire scene is as follows: Figure 4 As shown in Figure 2, the wind direction interpolation result for the target fire scene is as follows: Figure 5 shown.

[0058] The second object of the present invention is to provide a method and system for downscaling prediction of local wind field of forest fire considering fire-atmosphere interaction, comprising:

[0059] The acquisition module is used to obtain the time series data of meteorological attributes at different locations of the fire scene, the spatial location data of each meteorological station, and the time series data of infrared images of the fire spread process.

[0060] The calculation module is used to calculate the descriptive index of the fire spread situation according to the infrared image data of the fire spread process.

[0061] The key location meteorological data selection module is used to determine the spread direction of the fire head, fire tail and two wings according to the change of the temperature center of gravity during the fire spread process, and select the meteorological data near the fire head, fire tail and side wings according to the distance from the meteorological station to the straight line of each spread direction.

[0062] The prediction module is used to make predictions based on the four-directional meteorological data and the descriptive indicator data of the fire spread situation at the previous n moments, and obtain the prediction results of the four-directional wind speed and wind direction values ​​at the n+1th moment.

[0063] Interpolation module. It is used to interpolate the wind speed and wind direction values ​​of the four directions of the fire scene at the current moment, obtain the wind speed and wind direction values ​​of each position of the fire scene, and thus construct the local wind field of the fire scene.

Claims

1. A downscaling prediction method for local wind field of forest fire considering fire-atmosphere interaction, characterized in that: The method comprises: Step 1: Obtain the time series data of meteorological attributes at different locations of the fire scene, the spatial location of each meteorological station, and the time series data of infrared images of the fire spread process observed by remote sensing; Step 2: According to the direction of change of the temperature center of gravity during the fire spread process, the direction of the fire head is approximately determined, and then the direction of the fire tail and the two wings is determined; Step 3: Calculate the distance between the position of each weather station in step 1 and the straight lines in the four directions in step 2, and select the monitoring data of the weather station with the shortest distance to each straight line as the weather data near the fire head, the weather data near the fire tail, and the weather data near the two wings; Step 4: Extract the time series data of the descriptive indicators of the fire spread situation through the infrared image data of the fire spread process; Step 5: Based on the time series data of meteorological attributes of the four directions of the fire scene obtained in step 3 and the time series data of descriptive indicators of the fire spread situation obtained in step 4, the final data set is established; Step 6: Establish a deep learning model, and train a 4-direction wind speed and direction prediction model for the local forest fire based on the time series data set described in step 5; Step 7: Using the interpolation method, based on the output data of the 4-direction wind speed and direction prediction model of the fire scene, estimate the wind speed and direction at each location inside the fire scene to achieve wind field downscaling.

2. A downscaling prediction method for local wind field of forest fire considering fire-atmosphere interaction according to claim 1, characterized in that: The meteorological attributes described in step 1 include wind speed, wind direction, temperature, humidity, and air pressure; the meteorological stations described in step 1 should be evenly distributed around the fire scene, and the number of meteorological stations N>=4; the infrared image pixel value described in step 1 is equal to the temperature of the pixel point.

3. A downscaling prediction method for local wind field of forest fire considering fire-atmosphere interaction according to claim 2, characterized in that: The specific steps of step 2 are: Step 2.1: Binarize the infrared image using an adaptive temperature threshold, mark the area consisting of pixels with a temperature higher than the threshold as Burned, and mark the area consisting of pixels with a temperature lower than the threshold as Unburned; Step 2.2: Take the temperature values ​​of all pixels in the Burned area as the mass of the point and calculate the center of gravity (x, y) of the area; Step 2.3: Assume that the center of gravity of the fire scene infrared image at time t is (x t ,y t ), the center of gravity of the fire scene infrared image at time t+1 is (x t+1 ,y t+1 ), then the direction of the fire head at time t is represented by the vector (x t+1 -x t ,y t+1 -y t ) direction; the direction of the fire tail is in the opposite direction, and the directions of the two side wings form angles of 90° and 270° with the direction of the fire head respectively.

4. A downscaling prediction method for local wind field of forest fire considering fire-atmosphere interaction according to claim 3, characterized in that: The descriptive indicators of the fire spread situation described in step 4 include: the perimeter L of the burned area, the surface area S, the change in the horizontal coordinate of the center of gravity Δx, the change in the vertical coordinate Δy, and the elongation rate r of the perimeter L , area expansion rate r S ; r L , and r S The calculation formula is as follows: Where, L t+1 is the perimeter of the Burned area at time t+1; L t is the perimeter of the Burned area at time t; S t+1 is the area of ​​the Burned region at time t+1; S t is the area of ​​the Burned region at time t.

5. A downscaling prediction method for local wind field of forest fire considering fire-atmosphere interaction according to claim 4, characterized in that: The final data set described in step 5 is an M*N-order matrix, where M is the sum of the number of meteorological attribute variables and the descriptive indicators of the fire spread situation, and N is the number of samples, that is, the time dimension.

6. A downscaling prediction method for local wind field of forest fire considering fire-atmosphere interaction according to claim 5, characterized in that: The deep learning model described in step 6 is inputted with the meteorological attribute data of the four directions of the fire scene and the descriptive index data of the fire spread situation at the previous n moments, and the output is the wind speed and direction values ​​of the four directions at the n+1th moment.

7. A downscaling prediction method for local wind field of forest fire considering fire-atmosphere interaction according to claim 6, characterized in that: The deep learning model described in step 6 includes but is not limited to the CNN-LSTM model; the interpolation method described in step 7 includes but is not limited to the Kriging interpolation method.

8. A downscaling prediction system for local wind field of forest fire considering fire-atmosphere interaction, used to implement the downscaling prediction method for local wind field of forest fire considering fire-atmosphere interaction as described in any one of claims 1 to 7, characterized in that: include: The acquisition module is used to obtain the time series data of meteorological attributes at different locations of the fire scene, the spatial location data of each meteorological station, and the time series data of infrared images of the fire spread process; A calculation module, used for calculating the descriptive index of the fire spread situation according to the infrared image data of the fire spread process; The key position meteorological data selection module is used to determine the spread direction of the fire head, fire tail and two wings according to the change of the temperature center of gravity during the fire spread process, and select the meteorological data near the fire head, fire tail and flank according to the distance from the meteorological station to the straight line of each spread direction; The prediction module is used to make predictions based on the four-directional meteorological data and the descriptive index data of the fire spread situation at the previous n moments, and obtain the prediction results of the four-directional wind speed and wind direction values ​​at the n+1th moment; The interpolation module is used to interpolate the wind speed and wind direction values ​​of the four directions of the fire scene at the current moment to obtain the wind speed and wind direction values ​​of each position of the fire scene, thereby constructing the local wind field of the fire scene.

Images