Mountain fire risk prediction method, device and equipment under drought conditions

Abstract

The application discloses a mountain fire risk prediction method, device and equipment under drought conditions; the method comprises the following steps: preprocessing multi-source time series data of a target region, constructing a multi-dimensional feature sequence, performing drought prediction through a pre-trained drought prediction model, performing mountain fire risk prediction through a mountain fire risk prediction model based on the drought prediction result and the historical number of fire points, outputting a future mountain fire risk index and generating a mountain fire risk grade to generate early warning information; since the application realizes drought prediction by fusing multi-source time series data, historical drought evolution and atmospheric circulation information, and strengthens the description of mountain fire risk in an extreme drought scenario through an extreme drought nonlinear penalty factor, the accuracy and advance amount of the monthly prediction of mountain fire risk under drought conditions are improved, precise and reliable decision support for power grid disaster prevention and reduction is provided, and the influence risk of mountain fire under extreme drought weather conditions is reduced.

Description

Technical Field

[0001] This application relates to the field of wildfire risk prediction, and in particular to a method, apparatus and equipment for wildfire risk prediction under drought conditions. Background Technology

[0002] In recent years, wildfires have not only posed a serious threat to life, property, and the ecological environment, but have also easily caused power transmission line outages. This is especially true during extreme droughts, when wildfires have a particularly significant impact on power grid operation safety and power supply security. To effectively address the threat posed by wildfires under drought conditions, the power grid management system urgently needs to conduct quantitative forecasts of the risk of wildfires in the following month, one month in advance, so as to deploy firefighting equipment and reserve emergency resources such as water and fuel coal according to the risk level.

[0003] Currently, commonly used methods for wildfire risk prediction mainly focus on short-term risks over the next 3 to 7 days, typically based on empirical or statistical modeling of short-term meteorological factors, vegetation flammability, and fire weather indices. While these methods can provide risk warnings in a relatively short period, their predictive lead time is limited, making it difficult to support disaster prevention resource allocation decisions on a monthly scale.

[0004] The existing methods have the following main problems: First, short-term forecasts cannot meet the need to deploy emergency resources one month in advance; second, existing methods do not adequately characterize the dynamic evolution of drought and lack the technical means to transform meteorological elements into standardized drought indices and further predict future drought trends; third, existing methods do not fully consider the physical characteristics of the nonlinear and rapid amplification of wildfire risk under extreme drought conditions, making it difficult to accurately characterize the leap law of wildfire risk under extreme drought scenarios; fourth, existing models rarely combine the attention mechanism of atmospheric circulation factors with wildfire risk prediction, resulting in insufficient response capability of prediction results to circulation anomalies. Summary of the Invention

[0005] The main purpose of this application is to provide a method, device and equipment for wildfire risk prediction under drought conditions, which aims to solve the technical problems that the existing technology cannot meet the needs of long-term prediction in drought scenarios and is difficult to accurately capture the characteristics of wildfire risk in extreme drought scenarios.

[0006] To achieve the above objectives, this application provides a method for predicting wildfire risk under drought conditions, the method comprising the following steps: Acquire multi-source time-series data of the target area, and preprocess the multi-source time-series data to construct a multi-dimensional feature sequence. The multi-source time-series data includes time features, meteorological features, atmospheric circulation factors, and the number of historical fire points. Based on the meteorological characteristics, the drought index at each historical time step is calculated to generate a historical drought index sequence. The multidimensional feature sequence and the historical drought index sequence are input into a pre-trained drought prediction model to obtain drought prediction results, which include predicted future drought index values ​​and a global drought context vector. The drought prediction results and the number of historical fire spots are input into a pre-trained wildfire risk prediction model to obtain a future wildfire risk index. The wildfire risk prediction model is configured to calculate a basic wildfire risk value based on the drought prediction results and the number of historical fire spots, and to correct the basic wildfire risk value based on a preset extreme drought nonlinear penalty factor, and output the future wildfire risk index. Based on the future wildfire risk index, the wildfire risk level is determined, and wildfire risk warning information for the wildfire risk level is generated.

[0007] Furthermore, to achieve the above objectives, this application also proposes a wildfire risk prediction device under drought conditions that applies the wildfire risk prediction method under drought conditions described above. The wildfire risk prediction device under drought conditions includes: The data processing module is used to acquire multi-source time-series data of the target area, and to preprocess the multi-source time-series data to construct a multi-dimensional feature sequence. The multi-source time-series data includes time features, meteorological features, atmospheric circulation factors, and the number of historical fire points. The drought index calculation module is used to calculate the drought index at each historical time step based on the meteorological characteristics, and generate a historical drought index sequence. The drought prediction module is used to input the multidimensional feature sequence and the historical drought index sequence into a pre-trained drought prediction model to obtain drought prediction results, which include future drought index prediction values ​​and a global drought context vector. The wildfire risk prediction module is used to input the drought prediction results and the number of historical fire points into a pre-trained wildfire risk prediction model to obtain a future wildfire risk index. The wildfire risk prediction model is configured to calculate a basic wildfire risk value based on the drought prediction results and the number of historical fire points, and to correct the basic wildfire risk value based on a preset extreme drought nonlinear penalty factor, and output the future wildfire risk index. The wildfire early warning module is used to determine the wildfire risk level based on the future wildfire risk index and generate wildfire risk early warning information based on the wildfire risk level.

[0008] This application quantifies the dynamic process of drought evolution by calculating drought indices at historical time steps, transforming meteorological elements such as precipitation and evapotranspiration into standardized drought indicators. Through a drought prediction model, it performs time-series modeling and atmospheric circulation attention weighting on multidimensional feature sequences and historical drought index sequences, uncovering the long-term dependencies of drought development and the driving effect of circulation anomalies on drought, obtaining predicted future drought index values ​​and a global drought context vector. A wildfire risk prediction model integrates drought prediction results with historical fire point information and introduces an extreme drought nonlinear penalty factor to correct the basic risk, strengthening the exponential amplification characteristic of wildfire risk under extreme drought scenarios. Through risk level determination and early warning information generation, continuous risk indices are transformed into level information that can guide disaster prevention decisions. Therefore, this application, through time-series characterization of drought evolution, circulation attention-driven approaches, and extreme drought nonlinear correction, achieves quantitative prediction of wildfire risk under drought conditions on a monthly timescale, effectively solving the problems of insufficient short-term forecast lead and weak characterization of extreme drought risk, significantly improving the accuracy and interpretability of monthly wildfire risk prediction, and providing a reliable basis for early deployment of power grid disaster prevention. Attached Figure Description

[0009] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0010] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A flowchart illustrating the first embodiment of the wildfire risk prediction method under drought conditions provided in this application; Figure 2 A flowchart illustrating the second embodiment of the wildfire risk prediction method under drought conditions provided in this application; Figure 3 A flowchart illustrating the third embodiment of the wildfire risk prediction method under drought conditions provided in this application; Figure 4 A flowchart illustrating the fourth embodiment of the wildfire risk prediction method under drought conditions provided in this application; Figure 5 This is a schematic diagram of the module structure of the wildfire risk prediction device under drought conditions according to an embodiment of this application; Figure 6 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the wildfire risk prediction method under drought conditions in the embodiments of this application.

[0012] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0013] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0014] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0015] It should be noted that the executing entity of this application embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a computer or server, or an electronic device capable of performing the above functions. The following uses a wildfire risk prediction device under drought conditions (hereinafter referred to as the prediction device) as an example to illustrate this embodiment and the following embodiments.

[0016] Based on this, the first embodiment of this application provides a method for predicting wildfire risk under drought conditions, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the wildfire risk prediction method under drought conditions proposed in this application.

[0017] In this embodiment, the wildfire risk prediction method under drought conditions includes: Step S10: Obtain multi-source time-series data of the target region, and preprocess the multi-source time-series data to construct a multi-dimensional feature sequence.

[0018] It should be noted that multi-source time-series data includes time characteristics, meteorological characteristics, atmospheric circulation factors, and the number of historical fire points.

[0019] Meteorological characteristics refer to physical quantities that reflect the meteorological conditions of the target area, such as monthly cumulative precipitation, monthly average temperature, monthly average wind speed, and monthly average relative humidity. Atmospheric circulation factors refer to indices that describe the state of large-scale atmospheric circulation, such as the Southern Oscillation Index and the Western Pacific Subtropical High Area Index. The number of historical fire points refers to the total number of fire points detected by satellite in the target area over various time steps in the past.

[0020] In the specific implementation, the prediction device obtains the target area over the past n months (time index denoted as n). Using time-series data, a multidimensional feature matrix is ​​constructed, specifically including: (1) At time t (current month), extract the following data from the past n months: ①Time characteristics Months (January-December); ② Meteorological characteristics Monthly cumulative precipitation (Pi) and average monthly temperature Average monthly wind speed Monthly average relative humidity ; ③Atmospheric circulation factors Southern Oscillation Index, Western Pacific Subtropical High Area Index, etc.; ④ Historical fire points The number of lunar fire points detected by satellites in this region.

[0021] (2) Periodic mapping transformation of time features using sine and cosine functions: (3) Use Z-score normalization to eliminate the influence of dimensions (taking temperature as an example, the same applies to other quantities): in, This represents the standardized temperature value for the i-th month. This represents the actual average monthly temperature of the i-th month. This represents the average of historical temperature data for the target area. This represents the standard deviation of historical temperature data for the target area.

[0022] (4) All preprocessed features are concatenated into the input feature vector at time i. This yields an input sequence of length n: Step S20: Calculate the drought index for each historical time step based on the meteorological characteristics to generate a historical drought index sequence.

[0023] It should be noted that the drought index is a dimensionless indicator that quantifies the deviation of water surplus or deficit from the historical level at a certain time step. The smaller the value, the more severe the drought.

[0024] Understandably, by transforming meteorological characteristics into standardized drought indices, the historical drought evolution process is quantified, providing drought prediction models with input information that reflects the dynamics of water deficit.

[0025] In some embodiments, the forecasting device calculates the potential evapotranspiration using the evapotranspiration formula based on the cumulative precipitation, average temperature, maximum temperature, minimum temperature, and solar radiation at the top of the atmosphere at each time step, and obtains the water deficit based on the difference between the potential evapotranspiration and the cumulative precipitation. Furthermore, it calculates the standardized drought index based on the degree of deviation of the water deficit from the historical mean and standard deviation of the water deficit for the same month. The drought indices at each time step are arranged in chronological order to obtain the historical drought index sequence.

[0026] Step S30: Input the multidimensional feature sequence and the historical drought index sequence into the pre-trained drought prediction model to obtain drought prediction results.

[0027] It should be noted that drought prediction models refer to machine learning models used to predict future drought conditions based on historical multidimensional features and drought indices. Drought prediction results refer to the predicted information output by the drought prediction model, including predicted future drought index values ​​and a global drought context vector. The global drought context vector is a vector representation that integrates information from all historical time steps and highlights the influence of atmospheric circulation anomalies.

[0028] It is understandable that this embodiment models the evolution of drought through a time series model and introduces an atmospheric circulation attention mechanism, which can capture long-term dependencies and circulation anomalies, thereby improving the accuracy of future drought index predictions.

[0029] Step S40: Input the drought prediction results and the number of historical fire spots into the pre-trained wildfire risk prediction model to obtain the future wildfire risk index.

[0030] It should be noted that the wildfire risk prediction model is configured to calculate a basic wildfire risk value based on the drought prediction results and the number of historical fire points, and to correct the basic wildfire risk value based on a preset extreme drought nonlinear penalty factor, and output a future wildfire risk index.

[0031] It should be noted that the wildfire risk prediction model refers to a model used to predict the future wildfire risk index based on drought prediction results and historical fire point information. The basic wildfire risk value is a probability-based index of wildfire occurrence calculated under normal meteorological and drought conditions. The extreme drought nonlinear penalty factor is a correction coefficient that is activated and grows exponentially when the predicted drought index is lower than a preset extreme drought threshold. The future wildfire risk index is a quantitative indicator of future wildfire risk output after comprehensively considering the basic risk and the extreme drought penalty.

[0032] It is understandable that this embodiment links drought prediction with wildfire risk prediction and dynamically corrects the basic risk through an extreme drought nonlinear penalty factor, which can highlight the nonlinear leap characteristics of wildfire risk under extreme drought scenarios.

[0033] In some embodiments, the wildfire risk prediction model calculates a basic wildfire risk value based on the predicted future drought index, the global drought context vector, and the number of historical fire points, and determines whether the predicted future drought index is lower than a preset extreme drought physical threshold. If extreme drought is identified, a penalty factor is calculated based on the difference between the predicted future drought index and the threshold, along with a penalty adjustment coefficient. The penalty factor is then fused with the basic wildfire risk value to output the future wildfire risk index.

[0034] In some embodiments, the wildfire risk prediction model can be jointly trained with the drought prediction model, and a smooth loss function can be used to resist the interference of abnormal samples.

[0035] Step S50: Determine the wildfire risk level based on the future wildfire risk index, and generate wildfire risk warning information for the wildfire risk level.

[0036] It should be noted that fire risk warning information refers to information used to inform users of the current wildfire risk level and guide disaster prevention decisions; it converts continuous risk indices into discrete risk levels and generates warning information, which facilitates power grid operators to deploy emergency resources such as fire-fighting equipment, water and coal storage in advance according to the risk level.

[0037] In its implementation, the server compares the future wildfire risk index with multiple preset risk thresholds. For example, the risk index range of 0 to 100 is divided into four levels: low, medium, high, and extremely high, with corresponding ranges set to 0 to 30, 30 to 60, 60 to 85, and 85 to 100, respectively. The risk of wildfires is low; Moderate wildfire risk; High risk of wildfires; The risk of wildfires is extremely high.

[0038] When the risk index falls into the corresponding range, the wildfire risk level is determined and an early warning message containing the risk level, target area and forecast period is generated and sent to the dispatch terminal through the display interface or communication network.

[0039] This embodiment achieves quantitative prediction of wildfire risk under drought conditions on a monthly timescale by characterizing the drought evolution time series, driving circulation attention, and correcting for extreme drought nonlinearity. It effectively solves the problems of insufficient short-term forecast lead time and weak characterization of extreme drought risk, significantly improves the accuracy and interpretability of monthly wildfire risk prediction, and provides a reliable basis for early deployment of power grid disaster prevention.

[0040] Based on the first embodiment described above, in the second embodiment of this application, the same or similar content as the first embodiment can be referred to the above description, and will not be repeated hereafter.

[0041] Based on this, please refer to Figure 2 , Figure 2 This is a flowchart illustrating the second embodiment of the wildfire risk prediction method under drought conditions according to this application. In this embodiment, step S20 further includes: Step S201: Extract the cumulative precipitation, average temperature, maximum temperature and minimum temperature for each time step from the meteorological characteristics; Step S202: Obtain the latitude of the target area and the month corresponding to each time step, and determine the solar radiation at the top of the atmosphere corresponding to each time step based on the latitude and the month.

[0042] It should be noted that solar radiation at the top of the atmosphere refers to the solar radiation energy reaching the top of the Earth's atmosphere, used for evapotranspiration calculations. Latitude refers to the Earth's latitude coordinates of the target area.

[0043] In some embodiments, the server determines the amount of solar radiation at the top of the atmosphere at each time step by querying a pre-established table of solar radiation constants, based on the latitude of the center point of the target area and the month corresponding to each time step. This table of constants can be pre-constructed based on astronomical radiation formulas or meteorological observation data.

[0044] Step S203: Determine the potential evapotranspiration at each time step based on the solar radiation at the top of the atmosphere, the average temperature, the highest temperature, and the lowest temperature at each time step.

[0045] It should be noted that potential evapotranspiration refers to the maximum amount of water that the surface water may lose through evaporation and plant transpiration under certain meteorological conditions.

[0046] It should be understood that by combining solar radiation with temperature information, it is possible to estimate the water evaporation capacity of a target area at each time step, providing a key input for water deficit calculation.

[0047] In practical implementation, the potential evapotranspiration is calculated using the Hargreaves formula, referring to the following formula: in, Indicates the first Potential evapotranspiration (mm) for one month; indicating the first month. The solar radiation at the top of the atmosphere corresponding to the month; indicating the solar radiation at the top of the atmosphere for the month. The highest temperature (°C) of the month; indicating the month of The lowest temperature of the month (°C); indicating the month. The average monthly temperature (°C) for the month; This represents the preset Hargreaves empirical coefficient (e.g., 0.0023). This represents the empirical temperature correction constant (e.g., 17.8).

[0048] Step S204: Calculate the water deficit for each time step based on the cumulative precipitation and potential evapotranspiration at each time step.

[0049] It should be noted that water deficit refers to the difference between potential evapotranspiration and cumulative precipitation at a certain time step, reflecting the water surplus or deficit status at that time step.

[0050] It should be understood that by comparing precipitation and evapotranspiration, the water surplus or deficit status at each time step can be directly quantified. The water deficit is calculated using the following formula: in, Indicates the first Monthly water deficit (mm); Indicates the first The cumulative precipitation (mm) over the month.

[0051] Step S205: Calculate the drought index for each historical time step based on the water deficit at each time step, and generate a historical drought index sequence.

[0052] It should be noted that the historical monthly water deficit statistics refer to the statistical characteristics of the water deficit in the target area in the same historical period of each year, including the long-term average and standard deviation.

[0053] It should be understood that by standardizing the water deficit relative to historical statistics for the same month, seasonal differences are eliminated, making the drought severity in different months comparable. The drought index is calculated using the following formula: in, Indicates the first Standardized historical drought index for 1 month; This represents the average historical water deficit. This represents the standard deviation of historical water deficit.

[0054] This embodiment extracts meteorological elements, determines solar radiation, calculates potential evapotranspiration and water deficit, and standardizes the water deficit into a drought index, forming a complete calculation chain from raw meteorological data to standardized drought indicators. This enables the drought evolution process to be quantitatively characterized and provides reliable historical drought sequence input for subsequent monthly drought forecasting.

[0055] Based on the above embodiments, please refer to Figure 3 , Figure 3 This is a flowchart illustrating the third embodiment of the wildfire risk prediction method under drought conditions in this application. In this third embodiment, step S30 further includes: Step S301: Perform feature concatenation on the multidimensional feature sequence and the historical drought index sequence according to the time step to obtain the drought prediction input sequence; Step S302: Input the drought prediction input sequence into the pre-trained drought prediction model and output the drought prediction result; The drought prediction model includes a gated cyclic time-series feature extraction layer, an atmospheric circulation attention layer, and a drought prediction layer.

[0056] It should be noted that the gated recurrent temporal feature extraction layer refers to a neural network layer that uses a gated recurrent unit structure to extract features from temporal inputs. The atmospheric circulation attention layer refers to a neural network layer that calculates attention weights based on atmospheric circulation factors and performs weighted fusion of temporal features. The drought prediction layer refers to a fully connected network layer that maps the fused context vector to output a predicted value of the future drought index.

[0057] In the specific implementation, the drought prediction input sequence sequentially passes through a gated cyclic temporal feature extraction layer, an atmospheric circulation attention layer, and a drought prediction layer. The gated cyclic temporal feature extraction layer outputs the hidden state vector at each time step; the atmospheric circulation attention layer calculates the attention weights based on the atmospheric circulation factor vector and weights and fuses the hidden states to obtain the global drought context vector; the drought prediction layer performs a fully connected mapping on the global drought context vector and outputs the predicted value of the future drought index.

[0058] Furthermore, the gated cyclic temporal feature extraction layer is configured to extract features based on the drought prediction input sequence to obtain the hidden state vector at each time step; The atmospheric circulation attention layer is configured to extract atmospheric circulation factor vectors for each time step from the multidimensional feature sequence; determine the attention weights for each time step based on the hidden state vectors and the atmospheric circulation factor vectors; and perform weighted fusion of the hidden state vectors for each time step based on the attention weights for each time step to obtain a global drought context vector. The drought prediction layer is configured to perform a fully connected mapping on the global drought context vector to obtain a future drought index prediction value, and output the global drought context vector and the future drought index prediction value as the drought prediction result.

[0059] In practical implementation, the prediction device will input the drought prediction sequence. By time step Input-gated cyclic temporal feature extraction layer: in , Representing time steps Update gate and reset gate vector, Indicates time step The candidate hidden state vector, Indicates time step The output hidden state vector, Indicates the previous time step The hidden state vector, Indicates time step The input feature vector, , , These represent the weight matrices that the model can learn in the corresponding gating mechanism. , , These represent the bias vectors that the model can learn in the corresponding gating mechanism. This represents a vector concatenation operation. denoted by Sigmoid, and tanh by hyperbolic tangent activation function. This represents the Hadamard product (multiplication of corresponding matrix elements).

[0060] The processing steps for the atmospheric circulation attention layer and drought prediction layer are as follows: in Indicates time step Attention score vector, Indicates time step The eigenvectors of atmospheric circulation factors, , To compute the learnable weight matrix of the attention mechanism model, This is the bias vector that the model can learn when calculating the attention mechanism. Indicates time step The corresponding attention weight coefficients (scalars), where q is the attention query vector that the model can learn. Let exp be its transpose, representing the natural exponential function. This represents the global drought context vector that incorporates information from all time steps. This represents the predicted drought index value for the next month (time t+1) as output by the model. , These represent the weight matrix and bias vector of the fully connected prediction layer, respectively.

[0061] This embodiment extracts the temporal dependence of drought evolution through a gated recurrent unit, highlights the key impacts of circulation anomalies through an atmospheric circulation attention mechanism, and completes drought index regression through a fully connected layer. The three layers work together to enable the model to simultaneously capture the complex relationship between historical trends, circulation drivers, and future drought conditions, thereby improving the physical interpretability and prediction accuracy of drought prediction.

[0062] Based on the above embodiments, please refer to Figure 4 , Figure 4 This is a flowchart illustrating the fourth embodiment of the wildfire risk prediction method under drought conditions in this application. In this fourth embodiment, the wildfire risk prediction process of the wildfire risk prediction model specifically includes: Step S401: Calculate the average number of historical fire points according to the time step corresponding to the number of historical fire points to obtain the historical average number of fire points. Step S402: Standardize the historical average number of fire points to obtain historical fire point characterization values; Step S403: Perform feature weighting and logical activation on the predicted future drought index, the global drought context vector, and the historical fire point characterization value to obtain the basic wildfire risk value, referring to the following formula: in, This indicates the basic wildfire risk value; Use the Sigmoid activation function; This represents the learnable weight parameters corresponding to the predicted future drought index values; The learnable weight parameter represents the historical average number of fire points. This represents the learnable weight parameters corresponding to the global drought context vector; Indicates time step The predicted drought index value; Indicates the historical average number of fire points; Indicates the step up to the current time. The global drought context vector; representing the learnable bias parameters in the basic wildfire risk calculation process; Subscripts used to identify parameters for wildfire risk calculation.

[0063] This embodiment constructs a basic wildfire risk metric that comprehensively reflects the historical fire background, future drought trends, and global drought context by averaging historical fire points, standardization, and multi-source feature weighted activation, providing a reliable basic risk input for subsequent extreme drought correction.

[0064] Furthermore, to more accurately characterize the physical features that dramatically increase the risk of wildfires under extreme drought conditions, the steps following step S403 also include: Step S404: Compare the predicted future drought index with the preset extreme drought physical threshold to obtain the extreme drought determination result.

[0065] It should be noted that the preset extreme drought physical threshold refers to the critical drought index value used to determine whether an extreme drought state will occur in the future. The extreme drought determination result is the judgment conclusion obtained by comparing the predicted future drought index value with the preset extreme drought physical threshold. By comparing with the preset extreme drought physical threshold, it is possible to identify whether an extreme drought state will occur in the future.

[0066] Step S405: Determine the penalty activation state of the preset extreme drought nonlinear penalty factor based on the extreme drought determination result.

[0067] It should be noted that the penalty activation state refers to the state in which the nonlinear penalty factor for extreme drought is activated or deactivated. The penalty adjustment coefficient is a positive parameter that controls the steepness of risk escalation under extreme drought conditions. By determining the penalty activation state based on the extreme drought assessment results, unnecessary amplification of basic risks under non-extreme drought conditions can be avoided, ensuring the rationality of risk prediction.

[0068] Step S406: Based on the penalty activation state, the basic wildfire risk value is nonlinearly corrected using the preset extreme drought nonlinear penalty factor, and the correction result is constrained by the risk index range to output the future wildfire risk index.

[0069] It should be understood that nonlinear correction can characterize the exponential amplification effect of extreme drought on wildfire risk; range constraints can ensure that the future wildfire risk index falls within a uniform dimension range, facilitating risk classification comparison. The nonlinear correction process for the basic wildfire risk value refers to the following formula: in, This represents the nonlinear penalty factor for extreme drought. This indicates the preset physical threshold for extreme drought; This represents the penalty adjustment coefficient, which is a constant greater than 0, used to control the rate at which the penalty factor increases with the degree of drought. This indicates the future risk index of wildfires; The weighting coefficients representing the impact of extreme drought conditions on wildfire risk; This represents the upper limit of the normalized wildfire risk value; This is the scaling factor for the risk index.

[0070] This embodiment enables wildfire risk prediction to respond sensitively to extreme drought scenarios by using extreme drought determination, penalty activation state control, and nonlinear correction based on penalty factors. After the drought index exceeds the critical threshold, the risk index exhibits a nonlinear jump, thus more accurately depicting the physical characteristics of the sharp increase in wildfire risk under extreme drought conditions.

[0071] Based on the above embodiments, in the fifth embodiment of the wildfire risk prediction method under drought conditions in this application, the drought prediction model and the wildfire risk prediction model are jointly trained and optimized based on a joint loss function, which is: in, This represents the total loss function value during joint training; This represents the mean squared error loss value of the drought prediction model in performing the drought prediction task; This represents the smoothed L1 loss (Huber Loss) value of the wildfire risk prediction model when performing wildfire risk prediction tasks; This represents the loss weight hyperparameter for the wildfire risk prediction task, used to balance the impact of the drought index prediction task and the wildfire risk prediction task on model training.

[0072] The mean squared error loss value of the drought prediction model in performing drought prediction tasks is calculated using the following formula: The smoothed L1 loss value of the wildfire risk prediction model for performing wildfire risk prediction tasks is calculated using the following formula: in, This indicates the number of training samples contained in the current training batch, i.e., the batch size. This indicates the index of the training sample in the current training batch; This represents the predicted drought index value for the next time step corresponding to the kth training sample; This represents the actual drought index at the next time step corresponding to the k-th training sample; This represents the wildfire risk prediction value for the next time step corresponding to the k-th training sample; This represents the actual wildfire risk value at the next time step corresponding to the k-th training sample; This represents the Huber loss function (i.e., the L1 loss function) that can resist outlier interference.

[0073] In one example, a key forest area in Southwest China prone to forest fires is selected as the target region. The current time t is July. Based on observation data from the past n=3 months (i.e., May, June, and July, corresponding to time steps t-2, t-1, and t), the drought index and wildfire risk index for the next month (August, i.e., time t+1) need to be predicted. The specific steps include the following: Step 1: Acquisition and preprocessing of multi-source data.

[0074] The prediction device acquires time-series data of the target region over the past n months (time index denoted as i=t-n+1,…,t) and constructs a multidimensional feature matrix.

[0075] (1) Extract data from the target region from May to July. Taking July (time step t) as an example: ①Month: ; ② Meteorological characteristics: monthly cumulative precipitation mm, average monthly temperature ℃, average monthly wind speed m / s, monthly average relative humidity ; ③ Circulation factor: Southern Oscillation Index (ENSO) ; ④ Historical fire points: indivual.

[0076] (2) Periodic mapping transformation of time features using sine and cosine functions: (3) Use Z-score normalization to eliminate the influence of dimensions (taking temperature and humidity as examples): The historical average temperature for July was obtained by analyzing historical databases. ℃, standard deviation Calculate the standardized temperature .

[0077] The historical average temperature for July was obtained by analyzing historical databases. Standard deviation Calculate the standardized relative humidity .

[0078] (4) Similarly, the data (precipitation, wind speed, etc.) for May and June are processed and spliced ​​to obtain an input sequence containing the characteristics of arid climate. .

[0079] Step 2: Calculate the historical drought index.

[0080] (1) Substitute the highest, lowest, and average temperatures in July and the solar radiation constant from the table, and calculate the July temperature using the Hargreaves formula. mm.

[0081] (2) Calculate the water deficit and drought index: Water deficit mm.

[0082] Retrieve the average historical water deficit for the target area in July. Standard deviation Calculation Similarly, the drought index for May and June was calculated as follows: and .

[0083] Step 3: Predict the drought index for next month.

[0084] (1) GRU temporal feature extraction: Will include drought evolution ( → → ) characteristic sequence Input the GRU network to obtain the hidden state.

[0085] (2) Atmospheric circulation attention mechanism and drought prediction: The "Atmospheric Circulation Attention Mechanism" detected a continuously rising Southern Oscillation Index (SOI). ), calculate the attention weights for the July data. Reaching 0.65, generating drought context values. .

[0086] The predicted value is output after the fully connected layer: .

[0087] Step 4: Wildfire risk forecast for next month.

[0088] (1) Substitute the predicted drought index, historical fire points and other characteristics into the model to calculate the basic risk. .

[0089] (2) Constructing the nonlinear penalty factor for extreme drought and the final risk output: Set extreme drought threshold Penalty adjustment coefficient Extreme environment weighting coefficient Due to the predicted drought index in August. Triggering the penalty mechanism: Incorporate the penalty factor into the final risk calculation formula: Step 5: Model loss function calculation and joint training.

[0090] A training set is constructed using historical monthly data from the past 20 years (e.g., 2000-2023) for the target region. The batch size is set to N=64, and the loss weight β=0.8. The Adam optimizer is used to jointly minimize the MSE loss for drought prediction and the Huber loss for risk prediction. After 150 epochs of iteration, the model converges, and the weight parameters are saved for real-time use in the above steps.

[0091] Step 6: Risk classification and early warning.

[0092] Based on the wildfire risk index for next month (August) obtained in step four. Based on the assessment, 97.2 ∈ [85, 100], which is considered to be at extremely high risk of wildfire.

[0093] Based on the wildfire risk forecast results for August, the power grid should deploy firefighting equipment in advance and make preparations for water and coal storage.

[0094] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the wildfire risk prediction method under drought conditions proposed in this application. Any simple modifications based on this technical concept are within the scope of protection of this application.

[0095] This application also provides a wildfire risk prediction device under drought conditions; please refer to [reference needed]. Figure 5 , Figure 5 This is a schematic diagram of the module structure of a wildfire risk prediction device under drought conditions. The device includes: The data processing module 10 is used to acquire multi-source time-series data of the target area, and preprocess the multi-source time-series data to construct a multi-dimensional feature sequence. The multi-source time-series data includes time features, meteorological features, atmospheric circulation factors and the number of historical fire points. The drought index calculation module 20 is used to calculate the drought index at each historical time step based on the meteorological characteristics, and generate a historical drought index sequence. The drought prediction module 30 is used to input the multidimensional feature sequence and the historical drought index sequence into a pre-trained drought prediction model to obtain drought prediction results, which include future drought index prediction values ​​and a global drought context vector. The wildfire risk prediction module 40 is used to input the drought prediction results and the number of historical fire points into a pre-trained wildfire risk prediction model to obtain a future wildfire risk index. The wildfire risk prediction model is configured to calculate a basic wildfire risk value based on the drought prediction results and the number of historical fire points, and to correct the basic wildfire risk value based on a preset extreme drought nonlinear penalty factor, and output a future wildfire risk index. The wildfire early warning module 50 is used to determine the wildfire risk level based on the future wildfire risk index and generate wildfire risk early warning information based on the wildfire risk level.

[0096] This embodiment extracts meteorological elements, determines solar radiation, calculates potential evapotranspiration and water deficit, and standardizes the water deficit into a drought index, forming a complete calculation chain from raw meteorological data to standardized drought indicators. This enables the drought evolution process to be quantitatively characterized and provides reliable historical drought sequence input for subsequent monthly drought forecasting.

[0097] The wildfire risk prediction device under drought conditions provided in this application employs the wildfire risk prediction method under drought conditions described in the above embodiments, and can solve the technical problems to be addressed by the aforementioned wildfire risk prediction method under drought conditions. Compared with the prior art, the beneficial effects of the wildfire risk prediction device under drought conditions provided in this application are the same as those of the wildfire risk prediction method under drought conditions provided in the above embodiments, and other technical features in the wildfire risk prediction device under drought conditions are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0098] This application provides a wildfire risk prediction device under drought conditions, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the wildfire risk prediction method under drought conditions in any of the above embodiments.

[0099] The following is for reference. Figure 6 The diagram illustrates a structural schematic of a wildfire risk prediction device suitable for implementing embodiments of this application under drought conditions. The wildfire risk prediction device under drought conditions in this application embodiment can be a computing device such as a server, workstation, industrial control computer, or cloud computing node used to execute the above-described wildfire risk prediction method. It is mainly used to process, model, and predict multi-source time-series data of the target area to output wildfire risk early warning information. Figure 6 The wildfire risk prediction device shown under drought conditions is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0100] like Figure 6As shown, the wildfire risk prediction device under drought conditions may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to the program stored in ROM 1002 (Read Only Memory) or the program loaded from storage device 1003 into RAM 1004 (Random Access Memory), such as performing multi-source time-series data preprocessing, historical drought index calculation, drought prediction, wildfire risk prediction, and risk classification. RAM 1004 also stores various programs and data required for the operation of the wildfire risk prediction device under drought conditions, such as multidimensional feature sequences, historical drought index sequences, model parameters, drought prediction results, wildfire risk indices, and early warning information. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via bus 1005. I / O interface 1006 is also connected to bus 1005.

[0101] Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, a meteorological data interface, a satellite remote sensing data interface, a keyboard, a mouse, etc., wherein the meteorological data interface is used to receive meteorological observation data of the target area, and the satellite remote sensing data interface is used to receive historical fire point monitoring data; output devices 1008 including, for example, an LCD display, a speaker, an alarm indicator light, etc., for outputting wildfire risk levels and early warning information; storage devices 1003 including, for example, magnetic tape, a hard disk, a solid-state drive, etc., for persistently storing programs, trained model parameters, and historical data; and communication devices 1009 for data interaction with meteorological centers, satellite ground stations, power grid dispatching systems, or other operational terminals. Communication device 1009 allows the wildfire risk prediction equipment under drought conditions to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows wildfire risk prediction equipment under drought conditions with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or have alternatively.

[0102] Specifically, in some embodiments, the storage device 1003 or ROM 1002 may also store program modules corresponding to the above-described method claims, such as a data preprocessing module, a historical drought index calculation module, a drought prediction module, a wildfire risk prediction module, and a risk classification module. The processing device 1001 can load the above-described program modules into RAM 1004 and execute them to implement the corresponding wildfire risk prediction method steps.

[0103] The wildfire risk prediction device under drought conditions provided in this application employs the wildfire risk prediction method under drought conditions described in the above embodiments, and can solve the technical problems to be addressed by the aforementioned wildfire risk prediction method under drought conditions. Compared with the prior art, the beneficial effects of the wildfire risk prediction device under drought conditions provided in this application are the same as those of the wildfire risk prediction method under drought conditions provided in the above embodiments, and other technical features of the wildfire risk prediction device under drought conditions are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0104] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0105] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0106] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the wildfire risk prediction method under drought conditions in the above embodiments.

[0107] The computer-readable storage medium provided in this application can be a tangible medium capable of storing computer programs, such as a hard disk, random access memory, read-only memory, flash memory, optical storage device, or magnetic storage device. When the computer program is executed by a processor, it implements the above-described wildfire risk prediction method under drought conditions.

[0108] The aforementioned computer-readable storage medium may be included in a wildfire risk prediction device under drought conditions; or it may exist independently and not be assembled into a wildfire risk prediction device under drought conditions.

[0109] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the drought-condition wildfire risk prediction device, cause the drought-condition wildfire risk prediction device to perform the drought-condition wildfire risk prediction method as described in the above embodiments.

[0110] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including LAN (Local Area Network) or WAN (Wide Area Network)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0111] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0112] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described wildfire risk prediction method under drought conditions, and is capable of solving the technical problems to be solved by the above-described wildfire risk prediction method under drought conditions. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the wildfire risk prediction method under drought conditions provided in the above embodiments, and will not be repeated here.

[0113] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for predicting wildfire risk under drought conditions, characterized in that, The method includes: Acquire multi-source time-series data of the target area, and preprocess the multi-source time-series data to construct a multi-dimensional feature sequence. The multi-source time-series data includes time features, meteorological features, atmospheric circulation factors, and the number of historical fire points. Based on the meteorological characteristics, the drought index at each historical time step is calculated to generate a historical drought index sequence. The multidimensional feature sequence and the historical drought index sequence are input into a pre-trained drought prediction model to obtain drought prediction results, which include predicted future drought index values ​​and a global drought context vector. The drought prediction results and the number of historical fire spots are input into a pre-trained wildfire risk prediction model to obtain a future wildfire risk index. The wildfire risk prediction model is configured to calculate a basic wildfire risk value based on the drought prediction results and the number of historical fire spots, and to correct the basic wildfire risk value based on a preset extreme drought nonlinear penalty factor, and output the future wildfire risk index. Based on the future wildfire risk index, the wildfire risk level is determined, and wildfire risk warning information for the wildfire risk level is generated.

2. The method as described in claim 1, characterized in that, The process of calculating the drought index at each historical time step based on the meteorological characteristics to generate a historical drought index sequence includes: The cumulative precipitation, average temperature, maximum temperature, and minimum temperature at each time step are extracted from the meteorological characteristics. Obtain the latitude of the target area and the month corresponding to each time step, and determine the solar radiation at the top of the atmosphere corresponding to each time step based on the latitude and the month; Based on the solar radiation at the top of the atmosphere, the average temperature, the highest temperature, and the lowest temperature at each time step, the potential evapotranspiration at each time step is determined. The potential evapotranspiration is calculated using the following formula: in, Indicates the first Potential evapotranspiration in the first month; indicating the first month's potential evapotranspiration. The solar radiation at the top of the atmosphere corresponding to the month; indicating the solar radiation at the top of the atmosphere for the month. The highest temperature of the month; indicating the month. The lowest temperature of the month; indicating the first month. The average monthly temperature over the past few months; This represents the preset Hargreaves empirical coefficient; This represents the empirical temperature correction constant; The water deficit at each time step is calculated based on the cumulative precipitation and potential evapotranspiration at each time step, using the following formula: in, Indicates the first Monthly water deficit; Indicates the first The cumulative rainfall over the month; The historical drought index is calculated based on the water deficit at each time step, generating a historical drought index series. The drought index is calculated using the following formula: in, Indicates the first Standardized historical drought index for 1 month; This represents the average historical water deficit. This represents the standard deviation of historical water deficit.

3. The method as described in claim 1, characterized in that, The step of inputting the multidimensional feature sequence and the historical drought index sequence into a pre-trained drought prediction model to obtain drought prediction results includes: The multidimensional feature sequence and the historical drought index sequence are concatenated according to the time step to obtain the drought prediction input sequence; The drought prediction input sequence is input into a pre-trained drought prediction model, which outputs the drought prediction result. The drought prediction model includes a gated cyclic time-series feature extraction layer, an atmospheric circulation attention layer, and a drought prediction layer.

4. The method as described in claim 3, characterized in that, The gated recurrent temporal feature extraction layer is configured to extract features based on the drought prediction input sequence to obtain the hidden state vector at each time step; The atmospheric circulation attention layer is configured to extract atmospheric circulation factor vectors for each time step from the multidimensional feature sequence; and to determine the attention weights for each time step based on the hidden state vectors and the atmospheric circulation factor vectors for each time step. The hidden state vectors at each time step are weighted and fused based on the attention weights at each time step to obtain the global drought context vector. The drought prediction layer is configured to perform a fully connected mapping on the global drought context vector to obtain a future drought index prediction value, and output the global drought context vector and the future drought index prediction value as the drought prediction result.

5. The method as described in claim 1, characterized in that, The calculation of the basic wildfire risk value based on the drought prediction results and the number of historical fire spots includes: The average number of historical fire points is calculated by averaging the number of historical fire points according to the time step corresponding to the number of historical fire points. The historical average number of fire points is standardized to obtain the historical fire point characterization value; The basic wildfire risk value is obtained by performing feature weighting and logical activation on the predicted future drought index, the global drought context vector, and the historical fire point characterization value, as shown in the following formula: in, This indicates the basic wildfire risk value; Use the Sigmoid activation function; This represents the learnable weight parameters corresponding to the predicted future drought index values; The learnable weight parameter represents the historical average number of fire points. This represents the learnable weight parameters corresponding to the global drought context vector; Indicates time step The predicted drought index value; Indicates the historical average number of fire points; Indicates the step up to the current time. The global drought context vector; representing the learnable bias parameters in the basic wildfire risk calculation process; Subscripts used to identify parameters for wildfire risk calculation.

6. The method as described in claim 5, characterized in that, The process of correcting the baseline wildfire risk value based on a preset extreme drought nonlinear penalty factor and outputting a future wildfire risk index includes: The predicted future drought index is compared with a preset extreme drought physical threshold to obtain the extreme drought determination result; The activation state of the preset extreme drought nonlinear penalty factor is determined based on the extreme drought determination result; Based on the penalty activation state, the baseline wildfire risk value is nonlinearly corrected using the preset extreme drought nonlinear penalty factor, and the correction result is constrained by a risk index range to output the future wildfire risk index, as shown in the following formula: in, This represents the nonlinear penalty factor for extreme drought. This indicates the preset physical threshold for extreme drought; This represents the penalty adjustment coefficient, which is a constant greater than 0, used to control the rate at which the penalty factor increases with the degree of drought. This indicates the future risk index of wildfires; The weighting coefficients representing the impact of extreme drought conditions on wildfire risk; This represents the upper limit of the normalized wildfire risk value; This is the scaling factor for the risk index.

7. The method as described in claim 1, characterized in that, The drought prediction model and the wildfire risk prediction model are jointly trained and optimized based on a joint loss function, which is: in, This represents the total loss function value during joint training; This represents the mean squared error loss value of the drought prediction model in performing the drought prediction task; This represents the smoothed L1 loss value of the wildfire risk prediction model when performing wildfire risk prediction tasks. This represents the loss weight hyperparameter for the wildfire risk prediction task, used to balance the impact of the drought index prediction task and the wildfire risk prediction task on model training.

8. The method as described in claim 7, characterized in that, The mean squared error loss value of the drought prediction model in performing drought prediction tasks is calculated using the following formula: The smoothed L1 loss value of the wildfire risk prediction model for performing wildfire risk prediction tasks is calculated using the following formula: in, This indicates the number of training samples included in the current training batch; This indicates the index of the training sample in the current training batch; This represents the predicted drought index value for the next time step corresponding to the kth training sample; This represents the actual drought index at the next time step corresponding to the k-th training sample; This represents the wildfire risk prediction value for the next time step corresponding to the k-th training sample; This represents the actual wildfire risk value at the next time step corresponding to the k-th training sample; This represents the smoothed L1 loss function.

9. A wildfire risk prediction device under drought conditions that applies the wildfire risk prediction method under drought conditions according to any one of claims 1 to 8, characterized in that, The device includes: The data processing module is used to acquire multi-source time-series data of the target area, and to preprocess the multi-source time-series data to construct a multi-dimensional feature sequence. The multi-source time-series data includes time features, meteorological features, atmospheric circulation factors, and the number of historical fire points. The drought index calculation module is used to calculate the drought index at each historical time step based on the meteorological characteristics, and generate a historical drought index sequence. The drought prediction module is used to input the multidimensional feature sequence and the historical drought index sequence into a pre-trained drought prediction model to obtain drought prediction results, which include future drought index prediction values ​​and a global drought context vector. The wildfire risk prediction module is used to input the drought prediction results and the number of historical fire points into a pre-trained wildfire risk prediction model to obtain a future wildfire risk index. The wildfire risk prediction model is configured to calculate a basic wildfire risk value based on the drought prediction results and the number of historical fire points, and to correct the basic wildfire risk value based on a preset extreme drought nonlinear penalty factor, and output the future wildfire risk index. The wildfire early warning module is used to determine the wildfire risk level based on the future wildfire risk index and generate wildfire risk early warning information based on the wildfire risk level.

10. A wildfire risk prediction device under drought conditions, characterized in that, The wildfire risk prediction device under drought conditions includes: a memory, a processor, and a wildfire risk prediction program under drought conditions stored in the memory. The processor is used to run the wildfire risk prediction program under drought conditions, and the wildfire risk prediction program under drought conditions is configured to implement the wildfire risk prediction method under drought conditions as described in any one of claims 1 to 8.

Images