Forest fire prediction method and system based on XGBoost-Attention BiLSTM model

Abstract

The invention discloses a forest fire prediction method and system based on an XGBoost-Attention BiLSTM model, and the method comprises the steps: S1, building a forest fire prediction sample data set, S2, building the XGBoost-Attention BiLSTM model, and carrying out the model training and learning through the forest fire prediction sample data set; s3, other factor data processing modules extract topographic factor data and vegetation factor data by using the topographic data and the vegetation data, and the XGBoost layer outputs a wildfire occurrence probability through a sigmoid activation function; and S4, collecting meteorological data, topographic data and vegetation data containing spatio-temporal information in prediction time, obtaining historical meteorological data before the prediction time, inputting the historical meteorological data into the XGBoost-Attention BiLSTM model, and obtaining a wildfire occurrence probability. According to the method, the wildfire occurrence probability of each position point at the prediction time can be obtained, the early warning wildfire point position and the corresponding wildfire occurrence probability can be output, and the early warning wildfire point position can be conveniently and timely prevented and treated.

Description

Technical Field

[0001] The present invention relates to the field of forest fire prediction, and in particular to a forest fire prediction method and system based on an XGBoost-AttentionBiLSTM model. Background Art

[0002] As the main body of terrestrial ecosystems, forests contain thousands of material or non-material resources, which have direct or indirect effects on human survival, reproduction, economic and social development. Forests can not only conserve water, regulate climate and purify air, but also prevent wind and sand, reduce soil erosion and protect biodiversity. In addition, forests can not only provide social and economic resources such as timber resources, food resources and medicinal resources, but also provide many conveniences for human production and life. Forest fire is one of the important natural disturbance processes in forest ecosystems, which can help forest ecosystems transform nutrients and renew species, but it is also a major natural disaster that threatens forest resources. On average, hundreds of thousands of forest fires occur each year around the world, which not only damages the forest ecological environment, but also causes a large number of casualties and economic losses. Since the 1990s, the area of ​​forest fires in the United States has shown a clear upward trend. In 1996 alone, the area of ​​forests destroyed by forest fires reached 2.45 million hectares. In addition, as of early September 2021, according to statistics from the U.S. National Interdepartmental Fire Center, there were more than 43,400 wildfires in 2021, and the burned area exceeded 2 million hectares. In February 2009, a major wildfire lasted for more than a month in Australia, burning an area of ​​410,000 hectares, killing 273 people, and causing economic losses of $2 billion. This shows that forest fires have a significant impact on both personal safety and social economy.

[0003] Moreover, as the global climate warms, it not only affects some human production activities, but also has a certain impact on forest fires. Due to climate warming, the amount of forest combustibles increases. Studies have shown that as the temperature gradually rises, the volatile oil content of Scots pine increases by 39% compared to the normal content, and the resin content increases by 32%, resulting in a significant increase in its flammability. In addition, due to climate change, the increase in high temperature and dry weather has increased the evaporation of surface combustibles, and the number of lightning has increased with climate change. At the same time, climate warming has made forest fire behavior more complex and changeable, greatly increasing the difficulty of forest fire fighting, and posing a greater threat to social economy and people's safety. Therefore, in order to ensure the sustainable development of China's forest resources, large-scale and high-frequency forest fire early warning methods are urgently needed.

[0004] In recent years, with the rise of machine learning algorithms, more and more studies have tried to apply machine learning algorithms to predict forest fire risks, and found that machine learning algorithms have a relatively superior performance in forest fire prediction. Existing machine learning models are still shallow machine learning models, with limited capabilities in processing highly complex and high-dimensional data, and may not be able to fully capture all potential nonlinear relationships and deep-level features. Therefore, based on remote sensing data and artificial intelligence algorithms, the establishment of a forest fire risk warning prediction model can not only provide new ideas for forest fire prevention and control, realize forest fire warning, provide technical support for related work of local forestry departments, and reduce some losses caused by forest fires, as well as unnecessary casualties. Summary of the invention

[0005] The purpose of the present invention is to provide a forest fire prediction method and system based on an XGBoost-AttentionBiLSTM model, and a forest fire prediction sample data set including terrain data, vegetation data, historical meteorological data, predicted meteorological data, and wildfire data containing spatiotemporal information is constructed. The XGBoost-AttentionBiLSTM model uses the forest fire prediction sample data set for model training and learning, wherein the BiLSTM model extracts information from historical meteorological data in both the forward and reverse directions of the time series, and can better understand the lag effect in the meteorological data, facilitating accurate prediction; the attention mechanism module enhances the ability to capture long-term dependencies in time series data, and can also more flexibly allocate the weight of influence on the occurrence of forest fires; the fully connected layer can fuse the meteorological factor output results, predicted meteorological data, terrain factor data, vegetation factor data, and wildfire data, and the XGBoost layer performs binary classification processing and outputs the probability of wildfire occurrence using a sigmoid activation function.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] A forest fire prediction method based on an XGBoost-AttentionBiLSTM model, the method comprising:

[0008] S1. Construct a forest fire prediction sample data set. The data in the forest fire prediction sample data set are stored based on spatiotemporal coding association according to terrain data, vegetation data, historical meteorological data, predicted meteorological data, and wildfire data containing spatiotemporal information. The predicted meteorological data is the meteorological data at the prediction time point, and the historical meteorological data is the meteorological data before the corresponding prediction time point.

[0009] S2. Construct the XGBoost-AttentionBiLSTM model and use the forest fire prediction sample data set for model training and learning. The XGBoost-AttentionBiLSTM model includes the BiLSTM model, the attention mechanism module, other factor data processing modules, the fully connected layer and the XGBoost layer. The BiLSTM model extracts information from the historical meteorological data in the forward and reverse directions of the time series and obtains the output vector X n And the last hidden state query vector q, the score S is calculated according to the following formula n :

[0010] S n =q T *X n +b, where q T is the transpose of the query vector q, n represents the sequence number of the meteorological data in time series, and b is the regularization term;

[0011] The attention mechanism module combines the score S n And the output vector X n Capture relationships, assign weights, and fuse them to obtain output results;

[0012] S3, other factor data processing module uses terrain data and vegetation data to extract terrain factor data and vegetation factor data. The fully connected layer performs full connection training processing on the output result of step S2, predicted meteorological data, terrain factor data, vegetation factor data, and wildfire data and outputs them to the XGBoost layer. The XGBoost layer outputs the probability of wildfire occurrence through the sigmoid activation function.

[0013] S4. Collect meteorological data, terrain data, and vegetation data containing spatiotemporal information at the prediction time, obtain historical meteorological data before the prediction time, input them into the XGBoost-AttentionBiLSTM model, and obtain the probability of wildfire occurrence.

[0014] In order to better implement the present invention, the XGBoost-AttentionBiLSTM model obtains the probability of wildfire occurrence according to the spatial position; sets a probability threshold, extracts the spatial positions where the probability of wildfire occurrence is greater than the probability threshold as warning wildfire points, and then synchronously outputs the warning wildfire points and the corresponding wildfire probability or visualizes them in a geographic map.

[0015] Preferably, a fire risk grade classification standard is constructed, and the probability of wildfire occurrence at the obtained spatial location is graded according to the fire risk grade classification standard, and then visualized on a geographic map.

[0016] Preferably, the spatiotemporal information includes time information and space information, the space information is longitude and latitude coordinate information, and the spatial position is longitude and latitude coordinate information.

[0017] Preferably, points and areas that have been repeatedly identified as warning wildfires and / or points and areas where wildfires have occurred multiple times in history are set as key monitoring objects, and real-time data collection is performed on the key monitoring objects and input into the XGBoost-AttentionBiLSTM model to output the real-time wildfire occurrence probability of the key monitoring objects or visualize it in real time on a geographic map.

[0018] Preferably, in step S2, the BiLSTM model includes a forward LSTM processing unit and a reverse LSTM processing unit. The forward LSTM processing unit forward fuses all meteorological data to extract information according to the time series and inputs it into the reverse LSTM processing unit. The reverse LSTM processing unit reversely fuses all meteorological data to extract information according to the time series.

[0019] Preferably, the attention mechanism module processing includes the following method: first, the score S n Normalize and get the weight coefficient α n , the expression is as follows:

[0020] Where N represents the total number of serial numbers of historical meteorological data and forecast meteorological data;

[0021] Then the weight coefficient α n With the output vector X n The output vector C is obtained by multiplication, and the expression is as follows:

[0022] Preferably, the vegetation data is derived from two data products, MCD12Q1.061 and MCD43A4.061, the terrain data is derived from the SRTM3 global digital elevation model data set, and the historical meteorological data and forecast meteorological data are derived from the meteorological center.

[0023] A forest fire prediction system based on an XGBoost-AttentionBiLSTM model comprises an XGBoost-AttentionBiLSTM model, a forest fire prediction sample data set and a data acquisition module. The data in the forest fire prediction sample data set are stored based on temporal and spatial coding association according to terrain data, vegetation data, historical meteorological data, predicted meteorological data and wildfire data containing temporal and spatial information. The predicted meteorological data is the meteorological data at the prediction time point, and the historical meteorological data is the meteorological data before the corresponding prediction time point. The XGBoost-AttentionBiLSTM model uses the forest fire prediction sample data set for model training and learning. The XGBoost-AttentionBiLSTM model comprises a BiLSTM model, an attention mechanism module, other factor data processing modules, a fully connected layer and an XGBoost layer. The BiLSTM model extracts information from the historical meteorological data in both the forward and reverse directions of the time series and obtains an output vector X n And the last hidden state query vector q, the score S is calculated according to the following formula n :S n =q T *X n +b, where q T is the transpose of the query vector q, n represents the sequence number of the meteorological data in time series, and b is the regularization term; the attention mechanism module combines the score S n And the output vector X n Relationship capture, weight allocation and fusion are performed to obtain output results; other factor data processing modules use terrain data and vegetation data to extract terrain factor data and vegetation factor data, and the fully connected layer performs fully connected training processing on the output result of step S2, predicted meteorological data, terrain factor data, vegetation factor data, and wildfire data and outputs them to the XGBoost layer, and the XGBoost layer outputs the probability of wildfire occurrence through the sigmoid activation function; the data acquisition module is used to collect meteorological data, terrain data, and vegetation data containing spatiotemporal information at the prediction time, and obtain historical meteorological data before the prediction time, input into the XGBoost-AttentionBiLSTM model, and obtain the probability of wildfire occurrence.

[0024] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0025] (1) The present invention constructs a forest fire prediction sample data set containing terrain data, vegetation data, historical meteorological data, predicted meteorological data, and wildfire data containing spatiotemporal information. The XGBoost-AttentionBiLSTM model uses the forest fire prediction sample data set for model training and learning, wherein the BiLSTM model extracts information from the historical meteorological data in both the forward and reverse directions of the time series, which can better understand the lag effect in the meteorological data and facilitate accurate prediction; the attention mechanism module enhances the ability to capture long-term dependencies in the time series data, and can also more flexibly allocate the weight of the impact on the occurrence of forest fires; the fully connected layer can fuse the meteorological factor output results, predicted meteorological data, terrain factor data, vegetation factor data, and wildfire data; the XGBoost layer performs binary classification processing and uses the sigmoid activation function to output the probability of wildfire occurrence.

[0026] (2) After determining the prediction time, the present invention obtains the historical meteorological data before the prediction time and uses the meteorological data, terrain data, and vegetation data containing spatiotemporal information at the prediction time as the data source. The trained XGBoost-AttentionBiLSTM model uses the data source to predict the probability of wildfire occurrence.

[0027] (3) The present invention can obtain the probability of wildfire occurrence at the predicted time according to the spatial location, and can also output the warning wildfire point and the corresponding wildfire occurrence probability based on the probability threshold, so as to facilitate timely preventive treatment of the warning wildfire point. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A method flow chart of the forest fire prediction method of the present invention;

[0029] Figure 2 It is a schematic diagram of the principle of the XGBoost-AttentionBiLSTM model in the embodiment;

[0030] Figure 3 It is a principle structure diagram of the BiLSTM model in the embodiment;

[0031] Figure 4 It is a principle structure diagram of the attention mechanism module in the embodiment;

[0032] Figure 5 It is a comparison chart of ROC curves of the XGBoost-AttentionBiLSTM model in the embodiment and other models;

[0033] Figure 6 A forest fire risk level distribution map of a certain area is given as an example for the embodiment;

[0034] Figure 7For example, a wildfire location and regional distribution map of key monitoring objects in a certain area is provided as an example of an embodiment;

[0035] Figure 8 The early warning wildfire point distribution map of two areas is given as an example in the embodiment;

[0036] Fig. 9 The probability of occurrence of wildfire at a certain wildfire point in the embodiment is arranged in time series. DETAILED DESCRIPTION

[0037] The present invention is further described in detail below in conjunction with embodiments:

[0038] Example

[0039] like Figure 1 , Figure 2 As shown, a forest fire prediction method based on the XGBoost-AttentionBiLSTM model includes:

[0040] S1. Construct a forest fire prediction sample data set. The data in the forest fire prediction sample data set are stored based on spatiotemporal coding association according to terrain data, vegetation data, historical meteorological data, predicted meteorological data, and wildfire data (the wildfire data in this embodiment comes from remote sensing products such as MODIS and VIIRS. For example, VIIRS remote sensing products are used. VIIRS remote sensing products can provide location information, scope and intensity information of global wildfire activities) containing spatiotemporal information (spatiotemporal information includes time information and space information. The time information is time information accurate to minutes. The time information format is *** year ** month ** day ** hour ** minute, and the space information is longitude and latitude coordinate information). The predicted meteorological data is the meteorological data at the prediction time point, and the historical meteorological data is the meteorological data before the corresponding prediction time point. The vegetation data of this embodiment comes from two data products, MCD12Q1.061 and MCD43A4.061, which respectively collect surface cover types and normalized difference vegetation index (NDVI); download MCD12Q1.061 and MCD43A4.061 data from NASA's LAADSDAAC, and use HDF-EOS to GeoTIFF conversion tools (such as HEG Tool or GDAL) to convert HDF format data into GeoTIFF format, and use time series analysis methods to smooth the NDVI data to reduce the impact of noise and outliers; this embodiment can also integrate the US GFS numerical prediction model as a meteorological prediction factor data set, and the historical meteorological data comes from the fifth generation reanalysis data set (ERA5-Land) of the European Center for Medium-Term Weather Forecasts (ECMWF), and the data features used include surface net solar radiation, 2-meter air temperature, 2-meter dew point temperature, wind speed, surface temperature, etc. The vegetation data of this embodiment at least includes air temperature (or 2-meter air temperature), dew point temperature (or 2-meter dew point temperature), soil temperature (or surface temperature), wind speed, precipitation, net surface solar radiation, surface coverage, vegetation surface temperature, vegetation spontaneous combustion temperature, etc., wherein the vegetation spontaneous combustion temperature is set according to the vegetation type, and the vegetation surface temperature is obtained by inversion calculation of the surface temperature, vegetation moisture content and surface coverage, and the vegetation moisture content is set according to the vegetation type. The surface temperature of vegetation, the spontaneous combustion temperature of vegetation and the potential fire source are direct indicators reflecting the possibility of forest fire. In the past forest fire prediction research, the reanalysis data set was mainly used as a meteorological factor. However, due to the time lag of the reanalysis data, it is difficult to synchronize with the occurrence of the fire. Therefore, the present invention innovatively proposes to introduce meteorological prediction factors, and at the same time combines meteorological lag factors (the meteorological lag factors of the present invention are the key considerations for predicting meteorological data), terrain factors and vegetation factors to achieve forest fire prediction.The terrain data comes from the SRTM3 global digital elevation model dataset (spatial resolution is 30 meters). The SRTM3 global digital elevation model dataset is reprojected, clipped, and mosaicked in ArcGIS. The slope and aspect data of the terrain are calculated using the "slope" tool and "aspect" tool in ArcGIS. The terrain data at least includes elevation latitude and longitude information, slope data, and aspect data. The historical meteorological data and forecast meteorological data come from the Meteorological Center.

[0041] S2. Construct the XGBoost-AttentionBiLSTM model and use the forest fire prediction sample data set for model training and learning. The XGBoost-AttentionBiLSTM model includes the BiLSTM model (Bi-LSTM is used to capture the previous and next contextual information in the time series data to help understand the impact of the weather conditions in the past few days on the occurrence of fires), the attention mechanism module (the attention mechanism allows the model to focus on the most important part of the input data and determine which days' weather data are most critical for predicting fires), other factor data processing modules, the fully connected layer (the fully connected layer integrates information from the Bi-LSTM and the attention mechanism to understand the input data from a global perspective) and the XGBoost layer (the XGBoost layer is the final output layer to generate the final prediction results, using its powerful classification and regression capabilities). The BiLSTM model extracts information from the historical meteorological data in the forward and reverse directions of the time series and obtains the output vector X n and the last hidden state query vector q. In some embodiments, Figure 3 As shown, the BiLSTM model includes a forward LSTM processing unit (the forward LSTM processing unit includes several STM modules, Figure 3 The lower row in the middle contains a forward LSTM processing layer with multiple STM modules) and a reverse LSTM processing unit (the reverse LSTM processing unit includes several STM modules, Figure 3 The top row in the middle contains a reverse LSTM processing layer with multiple STM modules), and the forward LSTM processing unit forward fuses all meteorological data according to the time series to extract information and inputs it into the reverse LSTM processing unit (such as Figure 3 As shown, W1~W n For n historical meteorological data of n days before and after, the forward LSTM processing unit has n STM modules, and the n STM modules extract information corresponding to the n historical meteorological data one by one. The next STM module also receives the information output by the previous STM module and fuses it with the information extracted from the historical meteorological data corresponding to the next STM module and transmits it downward). The reverse LSTM processing unit reversely fuses all meteorological data extraction information according to the time series (such as Figure 3 As shown, W1~W nFor n historical meteorological data of n days before and after, the reverse LSTM processing unit has n STM modules, and the n STM modules extract information corresponding to the n historical meteorological data one by one. The next STM module also receives the information output by the previous STM module and fuses the information extracted by the next STM module corresponding to the historical meteorological data and the information transmitted by the STM module corresponding to the forward LSTM processing unit and transmits it downward). Each STM module of the reverse LSTM processing unit will get a hidden state (i.e., h1~h n ), the last hidden state h n is the query vector q, the output vector X of the BiLSTM model n will Figure 3 The vectors h1 to h1 output in n The BiLSTM model can effectively process information extraction from meteorological data. The BiLSTM model captures bidirectional information in the sequence through two LSTM module layers in two directions: one processes the natural order of the sequence (from front to back) and the other processes the reverse order (from back to front). By learning these complex spatiotemporal patterns, the BiLSTM model can better understand the lag effect in meteorological data and provide more accurate results.

[0042] The score S is calculated according to the following formula n :

[0043] S n =q T *X n +b, where q T is the transpose of the query vector q (the last hidden state query vector of the BiLSTM model), n represents the sequence number of the meteorological data in time series, and b is the regularization term; Figure 4 As shown, the input query vector q and the output vector X of the BiLSTM model n Calculate the score S n .

[0044] The attention mechanism module combines the score S n And the output vector X n Relationship capture, weight assignment and fusion are performed to obtain the output result. Figure 4As shown, after the BiLSTM model passes through the bidirectional long short-term memory network, the self-attention mechanism of the attention mechanism module is introduced; in the attention mechanism, by allocating weights, the attention to important parts and the discarding of unimportant parts are realized; the model's ability to capture long-term dependencies in time series data is enhanced, and the weights of the impact of different lagged meteorological factors on the occurrence of forest fires can be more flexibly allocated. In some embodiments, the attention mechanism module processing includes the following method: first, the score S n Normalize and get the weight coefficient α n (like Figure 4 As shown, the score S n Normalized), the expression is as follows:

[0045] Where N represents the total number of serial numbers of historical meteorological data and forecast meteorological data; then the weight coefficient α n With the output vector X n The product operation is performed to obtain the output vector C (as the output result of the attention mechanism module), which is expressed as follows:

[0046] S3, other factor data processing modules use terrain data and vegetation data to extract terrain factor data and vegetation factor data. The fully connected layer performs fully connected training processing on the output results of step S2, predicted meteorological data, terrain factor data, vegetation factor data, and wildfire data and outputs them to the XGBoost layer. The XGBoost layer outputs the probability of wildfire occurrence through the sigmoid activation function. In the XGBoost layer, the classification probability can be obtained by logically transforming the output of the decision tree; for the binary classification problem of forest fire, the model converts the decision tree output score into probability through the sigmoid activation function; the final forest fire probability output is performed based on the XGBoost module, and the regularization term in XGBoost is optimized to prevent the model from overfitting; at the same time, the second-order derivative is used to make the loss more accurate and improve the accuracy of the prediction. The present invention allows different types of features to be input into different layers respectively to ensure that each feature can be properly processed; through the integrated learning method, a powerful learner composed of multiple "weak learners" produces the final output, reduces the risk of overfitting, and improves the robustness and prediction accuracy of the model. The present invention can comprehensively capture and utilize the influence of multiple factors, thereby making more accurate forest fire predictions in complex environments.

[0047] S4. Collect meteorological data, terrain data, and vegetation data containing spatiotemporal information at the prediction time, obtain historical meteorological data before the prediction time, input them into the XGBoost-AttentionBiLSTM model, and obtain the probability of wildfire occurrence.

[0048] The XGBoost-AttentionBiLSTM model is a model creatively constructed for solving the technical problem of the present invention. This embodiment compares the model performance of the XGBoost-AttentionBiLSTM model of the present invention with logistic regression, random forest, and XGBoost. The evaluation results of all models in the test set are as follows:

[0049]

[0050] The XGBoost-AttentionBiLSTM model of the present invention performs well in all indicators, with an accuracy rate of 94.15%, a precision rate of 0.9229, a recall rate of 0.9586, and an F1 score of 0.9404. In order to more comprehensively evaluate the performance of the XGBoost-AttentionBiLSTM model of the present invention, the present invention not only introduces a variety of common machine learning models for comparative analysis, including XGBoost, Random Forest, and Logistic Regression, but also further evaluates each model through the ROC curve; evaluation indicators such as accuracy and recall rate are easily affected by the model segmentation threshold. Therefore, the use of the ROC curve can more objectively evaluate the overall performance of the model. Figure 5 As shown, the AUC value of the XGBoost-AttentionBiLSTM model of the present invention is the highest, reaching 0.9828, which is significantly higher than other models. The AUC values ​​of random forest and XGBoost are 0.9684 and 0.9670 respectively, while the AUC value of the logistic regression model is the lowest, which is only 0.7706. The higher the AUC value of the XGBoost-AttentionBiLSTM model of the present invention, the better the comprehensive performance of the model under different thresholds, and the better it can distinguish positive and negative samples, which further verifies the superiority of the XGBoost-AttentionBiLSTM model in the forest fire prediction task.

[0051] In some embodiments, the XGBoost-AttentionBiLSTM model obtains the probability of wildfire occurrence according to the spatial position (i.e., longitude and latitude coordinate information). The present invention also sets a probability threshold, extracts the spatial positions where the probability of wildfire occurrence is greater than the probability threshold as warning wildfire points, and then synchronously outputs the warning wildfire points and the corresponding wildfire occurrence probabilities or visualizes them in a geographic map. Preferably, the present invention sets the points and areas that have been repeatedly determined as warning wildfire points and areas and / or points and areas where wildfires have occurred multiple times in history as key monitoring objects, collects real-time data on the key monitoring objects and inputs them into the XGBoost-AttentionBiLSTM model, outputs the real-time wildfire occurrence probability of the key monitoring objects or visualizes them in real time in a geographic map. Figure 6 As shown, this embodiment takes the Yunnan-Guizhou-Sichuan region in January 2023 as an example, and takes January 20, 2023 as the prediction time. According to the method of the present invention, a visualization chart of the probability of wildfires in the Yunnan-Guizhou-Sichuan region on January 20, 2023 is obtained. From the distribution of forest fire points on that day, it can also be seen that many forest fires occurred in the central and southern regions of Yunnan Province. The verification results show that the high-risk areas predicted by the present invention are highly consistent with the actual locations of the fire points, showing a high sensitivity to forest fires, which further verifies the prediction ability of the model. This advantage stems from the innovation in the model structure design, especially the combination of the efficient feature selection ability of XGBoost and the time series data processing advantage of the Attention mechanism; in addition, the deep learning characteristics of the BiLSTM model of the present invention allow it to learn more complex information from a large amount of historical data and prediction data sets, thereby improving the accuracy of the prediction. The present invention not only improves the prediction performance of forest fires, but more importantly, they provide more reliable support for the early warning system of forest fires, which helps to take measures in advance to reduce fire losses and ensure the safety of forest resources.

[0052] In some embodiments, the present invention also constructs a fire risk level classification standard, and the wildfire occurrence probability of the obtained spatial location is classified according to the fire risk level classification standard, and then visualized in a geographic map. This embodiment takes the Yunnan-Guizhou-Sichuan region in January 2023 as an example. The fire risk level classification standard for the Yunnan-Guizhou-Sichuan region is as follows:

[0053]

[0054]

[0055] This embodiment uses ArcGIS Pro to classify the fire hazard level and obtains the forest fire risk level in the Yunnan, Guizhou and Sichuan regions as trademarks.

[0056] In some embodiments, the present invention sets the points and areas that have been repeatedly determined as warning wildfires or / and the points and areas where wildfires have occurred multiple times in history as key monitoring objects, collects real-time data on the key monitoring objects and inputs them into the XGBoost-AttentionBiLSTM model, and outputs the real-time wildfire occurrence probability of the key monitoring objects or visualizes them in real time on a geographic map (this embodiment takes Yunnan, Guizhou and Sichuan as examples, and the key monitoring objects are visualized in the geographic map as shown in FIG. Figure 7 shown).

[0057] A forest fire prediction system based on an XGBoost-AttentionBiLSTM model includes an XGBoost-AttentionBiLSTM model, a forest fire prediction sample data set, and a data acquisition module. The data in the forest fire prediction sample data set are stored based on spatiotemporal coding association according to terrain data, vegetation data, historical meteorological data, predicted meteorological data, and wildfire data containing spatiotemporal information. The predicted meteorological data is the meteorological data at the prediction time point, and the historical meteorological data is the meteorological data before the corresponding prediction time point. The XGBoost-AttentionBiLSTM model uses the forest fire prediction sample data set for model training and learning. The XGBoost-AttentionBiLSTM model includes a BiLSTM model, an attention mechanism module, other factor data processing modules, a fully connected layer, and an XGBoost layer. The BiLSTM model extracts information from the historical meteorological data in both the forward and reverse directions of the time series and obtains an output vector X n And the last hidden state query vector q, the score S is calculated according to the following formula n :S n =q T *X n +b, where q T is the transpose of the query vector q, n represents the sequence number of the meteorological data in time series, and b is the regularization term. The attention mechanism module combines the score S n And the output vector X n Relationship capture, weight assignment and fusion are performed to obtain the output result. The other factor data processing module uses terrain data and vegetation data to extract terrain factor data and vegetation factor data. The fully connected layer performs full connection training processing on the output result of step S2, predicted meteorological data, terrain factor data, vegetation factor data, and wildfire data and outputs them to the XGBoost layer. The XGBoost layer outputs the probability of wildfire occurrence through the sigmoid activation function. The data acquisition module is used to collect meteorological data, terrain data, and vegetation data containing spatiotemporal information at the prediction time, and obtain historical meteorological data before the prediction time to input into the XGBoost-AttentionBiLSTM model, and obtain the probability of wildfire occurrence.

[0058] In some embodiments, the XGBoost-AttentionBiLSTM model obtains the probability of wildfire occurrence according to the spatial position (i.e., longitude and latitude coordinate information). The present invention also sets a probability threshold, extracts the spatial position where the probability of wildfire occurrence is greater than the probability threshold as the warning wildfire point, and then synchronously outputs the warning wildfire point and the corresponding wildfire probability or visualizes it in a geographic map. Figure 8 As shown, this embodiment selects two study areas, Yuxi Mountain, Jiangchuan District, Yuxi City, Yunnan Province (April 11 to April 16, 2023) and Guanzhuang Village, Wenquan Street, Anning City, Yunnan Province (April 9 to April 25, 2023) for model testing. The model of the present invention extracts historical meteorological data from the previous 5 days to the time of occurrence, predicted meteorological data (the prediction time is any day or several days from April 11 to April 20, 2023), terrain data, and vegetation data. According to the method of the present invention, the probability of wildfire occurrence is obtained, and the probability threshold is set to 0.7. Fig. 9 The probability of wildfire occurrence at a certain location in Guanzhuang Village, Wenquan Street, Anning City, Yunnan Province from April 0 to April 25, 2023 is shown. The probability threshold of 0.7 was exceeded from April 11 to April 22, 2023, thus predicting the time interval of forest fires, such as Figure 8 As shown, this embodiment selects spatial locations that exceed the probability threshold in Yuxi Mountain, Jiangchuan District, Yuxi City, Yunnan Province and Guanzhuang Village, Wenquan Street, Anning City, Yunnan Province on a certain day as warning wildfire points, and visualizes them in a geographic map.

[0059] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A forest fire prediction method based on XGBoost-AttentionBiLSTM model, characterized by: The methods include: S1. Construct a forest fire prediction sample data set. The data in the forest fire prediction sample data set are stored based on spatiotemporal coding association according to terrain data, vegetation data, historical meteorological data, predicted meteorological data, and wildfire data containing spatiotemporal information. The predicted meteorological data is the meteorological data at the prediction time point, and the historical meteorological data is the meteorological data before the corresponding prediction time point. S2. Construct the XGBoost-AttentionBiLSTM model and use the forest fire prediction sample data set for model training and learning. The XGBoost-AttentionBiLSTM model includes the BiLSTM model, the attention mechanism module, other factor data processing modules, the fully connected layer and the XGBoost layer. The BiLSTM model extracts information from the historical meteorological data in the forward and reverse directions of the time series and obtains the output vector X n And the last hidden state query vector q, the score S is calculated according to the following formula n : S n =q T *X n +b, where q T is the transpose of the query vector q, n represents the sequence number of the meteorological data in time series, and b is the regularization term; The attention mechanism module combines the score S n And the output vector X n Capture relationships, assign weights, and fuse them to obtain output results; S3, other factor data processing module uses terrain data and vegetation data to extract terrain factor data and vegetation factor data. The fully connected layer performs full connection training processing on the output result of step S2, predicted meteorological data, terrain factor data, vegetation factor data, and wildfire data and outputs them to the XGBoost layer. The XGBoost layer outputs the probability of wildfire occurrence through the sigmoid activation function. S4. Collect meteorological data, terrain data, and vegetation data containing spatiotemporal information at the prediction time, obtain historical meteorological data before the prediction time, input them into the XGBoost-AttentionBiLSTM model, and obtain the probability of wildfire occurrence.

2. The forest fire prediction method based on the XGBoost-AttentionBiLSTM model according to claim 1, characterized in that: The XGBoost-AttentionBiLSTM model obtains the probability of wildfire occurrence according to spatial location; A probability threshold is set, and the spatial locations where the probability of wildfire occurrence is greater than the probability threshold are extracted as warning wildfire points. The warning wildfire points and the corresponding wildfire occurrence probabilities are then synchronously output or visualized on a geographic map.

3. The forest fire prediction method based on the XGBoost-AttentionBiLSTM model according to claim 2, characterized in that: A fire risk classification standard is constructed, and the probability of wildfire occurrence in the obtained spatial location is classified according to the fire risk classification standard, and then visualized on the geographic map.

4. The forest fire prediction method based on the XGBoost-AttentionBiLSTM model according to claim 2, characterized in that: The spatiotemporal information includes time information and space information, the space information is longitude and latitude coordinate information, and the spatial position is longitude and latitude coordinate information.

5. The forest fire prediction method based on the XGBoost-AttentionBiLSTM model according to claim 2, characterized in that: Points and areas that have been repeatedly identified as warning wildfires or / and points and areas where wildfires have occurred multiple times in history are set as key monitoring objects. Real-time data collection is performed on the key monitoring objects and input into the XGBoost-AttentionBiLSTM model to output the real-time wildfire occurrence probability of the key monitoring objects or visualize it in real time on a geographic map.

6. The forest fire prediction method based on the XGBoost-AttentionBiLSTM model according to claim 1, characterized in that: In step S2, the BiLSTM model includes a forward LSTM processing unit and a reverse LSTM processing unit. The forward LSTM processing unit forward fuses all meteorological data to extract information according to the time series and inputs it into the reverse LSTM processing unit. The reverse LSTM processing unit reversely fuses all meteorological data to extract information according to the time series.

7. The forest fire prediction method based on the XGBoost-AttentionBiLSTM model according to claim 1, characterized in that: The attention mechanism module processing includes the following methods: first, the score S n Normalize and get the weight coefficient α n , the expression is as follows: Where N represents the total number of serial numbers of historical meteorological data and forecast meteorological data; Then the weight coefficient α n With the output vector X n The output vector C is obtained by multiplication, and the expression is as follows:

8. The forest fire prediction method based on the XGBoost-AttentionBiLSTM model according to claim 1, characterized in that: The vegetation data is derived from two data products, MCD12Q1.061 and MCD43A4.061; the terrain data is derived from the SRTM3 global digital elevation model dataset; and the historical meteorological data and forecast meteorological data are derived from the meteorological center.

9. A forest fire prediction system based on the XGBoost-AttentionBiLSTM model, characterized by: The method comprises an XGBoost-AttentionBiLSTM model, a forest fire prediction sample data set and a data acquisition module. The data in the forest fire prediction sample data set are stored based on the temporal and spatial coding association according to the terrain data, vegetation data, historical meteorological data, predicted meteorological data and wildfire data containing temporal and spatial information. The predicted meteorological data is the meteorological data at the prediction time point, and the historical meteorological data is the meteorological data before the corresponding prediction time point. The XGBoost-AttentionBiLSTM model uses the forest fire prediction sample data set for model training and learning. The XGBoost-AttentionBiLSTM model comprises a BiLSTM model, an attention mechanism module, other factor data processing modules, a fully connected layer and an XGBoost layer. The BiLSTM model extracts information from the historical meteorological data in the forward and reverse directions of the time series and obtains an output vector X n And the last hidden state query vector q, the score S is calculated according to the following formula n :S n =q T *X n +b, where q T is the transpose of the query vector q, n represents the sequence number of the meteorological data in time series, and b is the regularization term; the attention mechanism module combines the score S n And the output vector X n Capture relationships, assign weights, and fuse them to obtain output results; The other factor data processing module uses terrain data and vegetation data to extract terrain factor data and vegetation factor data. The fully connected layer performs fully connected training processing on the output result of step S2, predicted meteorological data, terrain factor data, vegetation factor data, and wildfire data and outputs them to the XGBoost layer. The XGBoost layer outputs the probability of wildfire occurrence through the sigmoid activation function. The data acquisition module is used to collect meteorological data, terrain data, and vegetation data containing spatiotemporal information at the prediction time, and obtain historical meteorological data before the prediction time, input into the XGBoost-AttentionBiLSTM model, and obtain the probability of wildfire occurrence.

Images