Real-time deduction method and system for forest fire spread based on digital twin and dynamic parameter inversion

Abstract

This invention relates to the fields of wildfire monitoring, computer vision, and digital twin technology. It proposes a method and system for real-time wildfire spread simulation using digital twins and dynamic parameter inversion. The method dynamically inverts the surface combustible moisture content through white balance adaptive correction and HSV color feature extraction; maps physical camera parameters to a virtual camera; and uses a spatial ray casting algorithm and a digital elevation model to calculate the three-dimensional geographic coordinates of the fire point; it determines the wildfire spread direction by weighting the wind field vector and the maximum upslope vector in real time; it calculates the true slope of the dynamic path by taking points forward from the fire point as the starting point and using the grid accuracy of the digital elevation model as the step size; and it inputs the dynamic parameters into the wildfire spread model for simulation to generate the fire spread boundary. This invention significantly improves the physical realism and practical accuracy of wildfire spread simulation in complex environments through virtual-real fusion and physical correction mechanisms.

Description

Technical Field

[0001] This invention relates to the fields of wildfire monitoring and early warning, computer vision, and digital twin simulation technology. More specifically, it relates to a method and system for real-time extrapolation of wildfire spread using digital twins and dynamic parameter inversion. Background Technology

[0002] Wildfires are among the most destructive natural disasters, characterized by their suddenness, immense destructiveness, and difficulty in handling and relief. Wildfires not only burn vast amounts of forest resources and disrupt the ecological balance, but also directly threaten forest infrastructure and the safety of people's lives and property. Achieving early detection, early suppression, and early extinguishment of wildfires hinges on accurate assessment of the fire's spread trend.

[0003] Therefore, constructing a high-precision real-time wildfire spread simulation system has become an urgent need in the field of emergency command. To scientifically predict the development of fire boundaries, scholars both domestically and internationally have proposed various mathematical models for forest fire spread (such as the Rothermel model in the United States and the FWI model in Canada). In China, the Wang Zhengfei forest fire spread model, due to its simple parameter structure and strong adaptability to China's terrain and vegetation characteristics, has been widely used in forest fire prevention and command systems. The Wang Zhengfei forest fire spread model measures the forest fire spread speed... Defined as initial velocity The product of multiple environmental correction factors (i.e.) Although the Wang Zhengfei forest fire spread model is theoretically mature, in practical engineering applications, it is highly sensitive to the accuracy of its input parameters, particularly the moisture content coefficient. (Moisture content coefficient) and true slope correction coefficient The method of obtaining data has always been a bottleneck restricting the accuracy of its simulations, resulting in significant discrepancies between the simulation results and the actual fire scene. The specific technical bottlenecks are mainly reflected in the following two aspects: (1) Moisture content of surface combustibles The acquisition is delayed and has a large error. This is the parameter with the largest fluctuation and the most sensitive in Wang Zhengfei's model. Currently, the main methods for obtaining the surface combustible moisture content include: meteorological interpolation method: using temperature and humidity data from surrounding meteorological stations to estimate, but due to the large differences in microclimate in mountainous areas, single-point data cannot represent the entire fire area; satellite remote sensing method: using satellite multispectral data for inversion, but satellite revisit cycles are long (cannot be real-time), resolution is low, and it is difficult to penetrate dense canopy layers to monitor the actual surface combustibles (dried grass, fallen leaves, etc.); traditional visual method: although attempts have been made to use surveillance cameras to analyze vegetation color, complex lighting conditions in the field (dawn and dusk, shadows, weather changes) can lead to severe image color shifts, making the color-based inversion results extremely unstable and unable to meet the model's accuracy requirements.

[0004] (2) Correction coefficient for actual slope Computational distortion Highly sensitive to terrain, existing simulation systems typically employ static GIS analysis methods, which suffer from the following problems: Static slope error: Traditional methods directly read the inherent slope of the DEM raster (usually the maximum slope). However, under the combined influence of wind and terrain, the spread of wildfires often follows the direction of the vector resultant force of both (e.g., spreading along a sloping hillside). In this case, the "true slope" of the actual path traversed by the fire front is not equal to the "maximum slope" on the map, leading to distortion of the slope parameters input to the model; Location difficulties: To calculate the true slope, accurate 3D coordinates of the fire point are required. Existing monocular monitoring cameras lack depth information, making it difficult to achieve accurate 3D positioning of distant fire points at low cost, resulting in a lack of accurate starting points for subsequent path simulations. Moreover, in practice, high-quality forest fire spread data samples are extremely scarce (small sample problem), and the terrain and vegetation vary greatly across different regions, resulting in poor generalization ability of data-driven models, making them difficult to apply directly to unfamiliar forest areas lacking historical data.

[0005] In summary, how to eliminate ambient light interference to accurately invert the moisture content of combustibles, and how to achieve fire point location based on monocular vision and calculate the true slope of the dynamic path obtained by the inference method, are the key technical challenges to improving the accuracy of Wang Zhengfei's model in practical applications. Summary of the Invention

[0006] This invention aims to address the aforementioned deficiencies in existing technologies by providing a method and system for real-time wildfire spread prediction using digital twins and dynamic parameter inversion. With wildfire monitoring and early warning as its core application area, it integrates the techniques of computer vision and digital twin simulation. This method can effectively eliminate ambient light interference to accurately and in real-time invert the moisture content of combustibles, and can achieve monocular passive positioning and calculate the true slope of dynamic paths based on digital twin technology.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A real-time wildfire spread prediction system based on digital twin and dynamic parameter inversion includes a multi-source sensing module, a dynamic inversion module, a digital twin computing module, and a prediction visualization module. The multi-source sensing module is integrated with a standard optical calibration target at a field monitoring station to collect visible light video streams, gimbal attitude data, and micro-meteorological data. The multi-source sensing module outputs the collected raw data to the dynamic inversion module and the digital twin computing module, respectively. The dynamic inversion module establishes an active calibration mechanism, acquires standard optical calibration target images and white balance correction matrices, performs color restoration through the white balance correction matrix, and outputs the relative moisture content of surface combustibles based on the mean hue and mean saturation in the HSV color space through a preset color-moisture content mapping model. The digital twin computing module has a built-in three-dimensional digital elevation model and a digital twin scene, which completes spatial ray projection positioning, propagation direction vector synthesis and real slope calculation. The outputs of the dynamic inversion module and the digital twin computing module are both transmitted to the simulation visualization module as input data, providing core input parameters for the simulation visualization module. The simulation visualization module calculates the wildfire spread rate based on the relative moisture content and the actual slope, and dynamically renders the fire boundary on the 3D map in combination with the spread direction.

[0008] A real-time wildfire spread prediction method based on digital twins and dynamic parameter inversion, using the aforementioned real-time wildfire spread prediction system based on digital twins and dynamic parameter inversion, eliminates ambient light interference by setting constant optical references at field monitoring stations, improves slope coordinate accuracy using digital twin scenes, determines the spread direction based on vector synthesis, performs path sampling with the accuracy of the digital elevation model as the step size, and calculates the true slope, significantly improving the input error of Wang Zhengfei's wildfire spread model. The method specifically includes the following steps: Step 1. Based on the white balance adaptive correction algorithm, perform color correction on the monitoring video stream acquired by the physical camera and the image of the preset standard optical calibration target, extract the optical features of the surface combustible area and invert the relative water content; Step 2. Synchronously map the real-time attitude parameters and zoom parameters of the physical camera to the virtual camera in the digital twin scene, so that the virtual camera and the physical camera maintain a completely consistent virtual and real attitude; emit a spatial ray from the viewpoint of the virtual camera to the pixel coordinates of the fire point target identified in the video image, and use the spatial ray projection algorithm to detect the first collision point between the emitted spatial ray and the three-dimensional digital elevation model mesh built into the digital twin scene, complete the calculation of the three-dimensional geographic coordinates of the fire point, obtain the ground projection of the surface normal vector at the corresponding location of the fire point to obtain the maximum upslope vector of the terrain, and then combine it with the wind field vector collected in real time by the wind direction and wind force sensors, and calculate the spread direction of the wildfire through vector weighted synthesis; Step 3. Starting from the fire point, and using the grid accuracy of the digital elevation model as the step size, take points forward along the spread direction obtained in Step 2, and calculate the true slope of the dynamic path. Step 4. Using the relative moisture content obtained in Step 1 and the actual slope obtained in Step 3 as dynamic parameters, input the preset Wang Zhengfei forest fire spread model to calculate the forest fire spread rate. Combine the forest fire spread rate with the spread direction determined in Step 2 to deduce the development trend of the fire boundary.

[0009] Furthermore, Wang Zhengfei's forest fire spread model uses the following formula to calculate the spread rate: in, For the speed of spread, The initial spread rate, This is the moisture content correction factor obtained from the inversion. This is the wind speed correction factor. This is the corrected coefficient for the actual slope obtained from the solution.

[0010] Furthermore, in step 1, the inversion of relative moisture content includes the following steps: Step 1.1 Preset a standard optical calibration target with constant optical properties within the field of view of the physical camera or on the pan-tilt support structure, establish an active calibration mechanism, control the pan-tilt unit where the camera is located to rotate at regular intervals, and the physical camera periodically aligns with the standard optical calibration target to acquire image data of the standard optical calibration target, calculate the RGB gain deviation under the current lighting environment to generate a white balance correction matrix; and correct the white balance correction matrix based on the acquired image data. Step 1.2 Use semantic segmentation algorithm to extract the area of ​​combustible material on the ground in the surveillance video stream, and apply the white balance correction matrix obtained in Step 1.1 to perform normalized color restoration; Step 1.3 Calculate the mean hue and mean saturation of the corrected surface combustible area in the HSV color space, and output the relative moisture content of the surface combustible through the preset color-moisture content mapping model.

[0011] Furthermore, in step 2, calculating the three-dimensional geographic coordinates of the fire point specifically includes the following steps: Step 2.1 Read the azimuth, pitch and zoom parameters of the camera in real time, convert the zoom parameters into the vertical field of view, and map the camera parameters to the virtual camera in the digital twin scene in real time to achieve virtual and real posture synchronization. Step 2.2 Use visual detection algorithms to locate the fire area in the surveillance video stream, draw a detection box based on the identified target, and extract the center point at the bottom of the detection box as the target pixel coordinates; Step 2.3: Launch a spatial ray from the virtual camera viewpoint toward the target pixel coordinates, and detect the first collision point between the spatial ray and the digital elevation model grid. The coordinates of the first collision point are the three-dimensional geographic coordinates of the fire point.

[0012] Furthermore, in step 2, the calculation of the wildfire spread direction follows the following vector composition formula: in, The propagation direction vector, The wind field vector, It is the vector in the opposite direction of the projection of the surface normal onto the horizontal plane, i.e., the vector of the maximum upslope of the terrain; and These are the weighting coefficients.

[0013] Furthermore, in step 3, calculating the actual physical slope includes the following steps: Step 3.1 Using the three-dimensional geographic coordinates of the fire point as the sampling starting point, determine the sampling endpoint by moving forward along the spread direction vector with the grid accuracy of the digital elevation model as the step size; Step 3.2 Calculate the horizontal projection distance and vertical elevation difference between the sampling start point and the sampling end point, and use the inverse trigonometric function to solve the true slope of the road segment between the sampling start point and the sampling end point.

[0014] In summary, compared with the prior art, the present invention has the following beneficial effects: To address the shortcomings of existing traditional visual methods, such as severe image color distortion and unstable inversion under complex lighting conditions in the field, this invention introduces a "standard optical calibration target + active calibration mechanism." By integrating and deploying constant optical reference objects at field monitoring stations, it effectively eliminates ambient light interference caused by dawn / dusk and shadows, achieving high-precision dynamic monitoring of surface combustible water content and solving the problems in Wang Zhengfei's model. The invention addresses the challenges of delayed and inaccurate parameter acquisition. Recognizing the limitations of conventional monocular cameras ("lacking depth information and difficult to locate") and lidar ("expensive and difficult to widely adopt"), this invention utilizes a spatial ray projection algorithm within a digital twin scenario to solve the pain point of long-distance 3D fire point localization. It achieves accurate passive fire point localization without requiring expensive additional hardware, providing a precise coordinate reference for subsequent slope calculations.

[0015] This invention eliminates static map errors and restores the true physical spread process: Addressing the slope calculation distortion caused by directly reading the inherent slope of the map in traditional static GIS methods, this invention determines the spread direction based on "vector synthesis" and calculates the true slope by sampling the path using the accuracy of the digital elevation model as the step size. This accurately restores the actual climbing path of the fire head under the combined effects of wind and terrain, fundamentally correcting the true slope correction coefficient in Wang Zhengfei's forest fire spread model. The input error of the parameters significantly improved the inference accuracy of Wang Zhengfei's forest fire spread model in complex terrain.

[0016] This invention deeply integrates multi-source sensing data, computer vision, and digital twin technology to construct an automated closed loop of "perception-inversion-deduction," solving the problems of fragmented data and slow manual judgment in traditional command and providing a powerful auxiliary decision-making tool for forest fire prevention. Attached Figure Description

[0017] Figure 1 This is an overall flowchart of the method of the present invention.

[0018] Figure 2 This is a flowchart for the dynamic inversion of surface combustible moisture content.

[0019] Figure 3 This is a schematic diagram illustrating the principle of fire point localization and real slope calculation based on digital twins.

[0020] Figure 4 This is a structural diagram of the real-time wildfire spread simulation system module of the present invention.

[0021] Figure 5 This is a schematic diagram of the data flow and interaction logic of the real-time wildfire spread simulation system of the present invention. Detailed Implementation

[0022] The present invention will now be described in further detail with reference to the accompanying drawings.

[0023] It should be noted that, for ease of description, the descriptions of direction in the following text are consistent with the directions in the accompanying drawings, but they do not limit the structure of the present invention.

[0024] like Figures 1-5 As shown, this invention discloses a real-time wildfire spread prediction system based on digital twin and dynamic parameter inversion, comprising a multi-source sensing module, a dynamic inversion module, a digital twin computing module, and a prediction visualization module. The multi-source sensing module is integrated with a standard optical calibration target at a field monitoring station. The monitoring station integrates and deploys the standard optical calibration target to collect visible light video streams, gimbal attitude data, and micro-meteorological data. The multi-source sensing module outputs the collected raw data to the dynamic inversion module and the digital twin computing module, respectively.

[0025] The dynamic inversion module establishes an active calibration mechanism, acquires standard optical calibration target images and white balance correction matrices, performs color restoration through the white balance correction matrix, and outputs the relative moisture content of surface combustibles based on the mean hue and mean saturation in the HSV color space through a preset color-moisture content mapping model.

[0026] The digital twin computing module has a built-in three-dimensional digital elevation model and a digital twin scene, which completes spatial ray projection positioning, propagation direction vector synthesis and real slope calculation.

[0027] The outputs of the dynamic inversion module and the digital twin computing module are both transmitted as input data to the simulation visualization module, providing core input parameters for the simulation visualization module. The simulation visualization module calculates the wildfire spread rate based on the relative moisture content and the actual slope, and dynamically renders the fire boundary on the three-dimensional map in combination with the spread direction.

[0028] This invention also discloses a real-time wildfire spread prediction method based on digital twins and dynamic parameter inversion. Based on the aforementioned real-time wildfire spread prediction system using digital twins and dynamic parameter inversion, it eliminates ambient light interference by setting constant optical references at field monitoring stations, improves slope coordinate accuracy using digital twin scenes, determines the spread direction based on vector synthesis, performs path sampling with the accuracy of the digital elevation model as the step size, and calculates the true slope, significantly improving the input error of Wang Zhengfei's wildfire spread model. Specifically, it includes the following steps: Step 1. Based on the white balance adaptive correction algorithm, perform color correction on the monitoring video stream acquired by the physical camera and the image of the preset standard optical calibration target, extract the optical features of the surface combustible area and invert the relative water content, including the following steps: Step 1.1 Pre-set a standard optical calibration target with constant optical properties within the field of view of the physical camera or on the pan-tilt support structure, establish an active calibration mechanism, control the pan-tilt unit where the camera is located to rotate at regular intervals, and the physical camera periodically aligns with the standard optical calibration target to acquire image data of the standard optical calibration target. Calculate the RGB gain deviation under the current lighting environment to generate a white balance correction matrix. Use the monitoring terminal to acquire the video stream, and combine it with the standard optical calibration target integrated and deployed on the field monitoring station to perform adaptive white balance correction to eliminate ambient light interference caused by dawn / dusk and weather changes. Correct the white balance correction matrix according to the acquired image data. The correction cycle can be preset according to actual needs. The white balance correction matrix remains unchanged after the previous correction is completed and before the next correction begins.

[0029] Step 1.2 Use a semantic segmentation algorithm to extract the area of ​​combustible materials on the ground in the surveillance video stream, and apply the white balance correction matrix obtained in Step 1.1 to perform normalized color restoration.

[0030] Step 1.3 Calculate the mean hue and mean saturation of the corrected surface combustible area in the HSV color space, and output the relative moisture content of the surface combustible through a preset color-moisture content mapping model. The color-moisture content mapping model is a binary nonlinear regression model with the mean hue and mean saturation as dual inputs. It is constructed by fitting the HSV color characteristics of combustibles with calibration samples of measured moisture content by the drying method using the least squares method. The inputs are the mean hue and mean saturation, and the output is the relative moisture content from 0 to 100%.

[0031] Step 2. Synchronously map the real-time attitude and zoom parameters of the physical camera to the virtual camera in the digital twin scene, ensuring complete consistency between the virtual and physical camera attitudes. Acquire the azimuth, pitch, and zoom parameters of the physical camera in real time and convert them into a vertical field of view to drive the virtual camera in the digital twin scene to maintain attitude synchronization. From the virtual camera's viewpoint, emit a spatial ray to the pixel coordinates of the identified fire point in the video image. This ray emission is completed within the digital twin scene, obtaining the geographical coordinates of the fire point and the surface normal vector. Projecting the normal vector onto the ground reveals the direction of the maximum slope. This is achieved by considering factors such as wind speed in the real scene. The wind direction sensor obtains the wind field vector, which is then weighted and synthesized with the maximum slope direction vector to determine the spread direction of the wildfire. This process is achieved through a ray casting algorithm using cameras and a digital twin, eliminating the need for expensive lidar and reducing detection costs. The spatial ray casting algorithm detects the first collision point between the emitted spatial ray and the 3D digital elevation model mesh built into the digital twin scene, completing the calculation of the 3D geographic coordinates of the fire point. The ground projection of the surface normal vector at the corresponding location of the fire point yields the maximum upslope vector. This, combined with the wind field vector collected in real time by wind direction and wind force sensors, is used to calculate the spread direction of the wildfire through vector weighting. The specific steps for calculating the 3D geographic coordinates of the fire point include: Step 2.1 Read the azimuth, pitch and zoom parameters of the camera in real time, convert the zoom parameters into the vertical field of view, and map the camera parameters to the virtual camera in the digital twin scene in real time to achieve virtual and real posture synchronization. Step 2.2 Use visual detection algorithms to locate the fire area in the monitoring video stream, draw a detection box based on the identified target, and extract the center point at the bottom of the detection box as the target pixel coordinates; YOLO series target detection algorithms such as YOLOv5 and YOLOv8, which can locate the fire area, can all achieve this.

[0032] Step 2.3: Launch a spatial ray from the virtual camera viewpoint toward the target pixel coordinates, and detect the first collision point between the spatial ray and the digital elevation model grid. The coordinates of the first collision point are the three-dimensional geographic coordinates of the fire point.

[0033] The following vector composition formula is used to calculate the direction of wildfire spread: in, The propagation direction vector, The wind field vector, It is the vector in the opposite direction of the projection of the surface normal onto the horizontal plane, i.e., the vector of the maximum upslope of the terrain; and These are the weighting coefficients.

[0034] Step 3. Starting from the fire point and using the grid accuracy of the digital elevation model as the step size, take points forward along the spread direction obtained in Step 2, calculate the true slope of the dynamic path, calculate the horizontal projection distance and vertical elevation difference between the sampling start and end points, and use inverse trigonometric functions to solve for the true slope on the dynamic path, including the following steps: Step 3.1 Using the three-dimensional geographic coordinates of the fire point as the sampling starting point, the sampling endpoint is determined by moving forward along the spread direction vector with the grid accuracy of the digital elevation model as the step size; in this embodiment, the step size is 30 meters, which is the DEM grid accuracy. In practical applications, the step size can be adjusted according to the requirements.

[0035] Step 3.2 Calculate the horizontal projection distance and vertical elevation difference between the sampling start point and the sampling end point, and use the inverse trigonometric function to solve the true slope of the road segment between the sampling start point and the sampling end point.

[0036] Step 4. Using the relative moisture content obtained in Step 1 and the actual slope obtained in Step 3 as dynamic parameters, input the preset Wang Zhengfei forest fire spread model to calculate the forest fire spread rate. Combine the forest fire spread rate with the spread direction determined in Step 2 to deduce the development trend of the fire boundary.

[0037] The forest fire spread model specifically adopts Wang Zhengfei's forest fire spread model, and its spread rate calculation formula is as follows: in, For the speed of spread, The initial spread rate, It is the moisture content correction factor obtained from the inversion. This is the wind speed correction factor. It is the actual slope correction coefficient obtained from step S4.

[0038] The digital twin computing module of this invention configures the parameters of the physical camera and the virtual camera to be consistent, ensuring that the changes in the real environment and the virtual environment are identical. The acquired parameters are correlated with the parameters of the physical camera, accurately reflecting changes in the real environment. By calculating the resultant force of wind field vectors and terrain vectors and sampling along a dynamic path, the true slope is calculated for correction. No historical datasets are required for training, and it is not limited by region or sample size, exhibiting stronger universality and physical interpretability. Through virtual-real fusion and physical correction mechanisms, this invention significantly improves the physical realism and practical accuracy of wildfire spread simulation in complex environments.

[0039] Example: like Figure 1 As shown, this embodiment relies on the collaborative operation of a field monitoring station and a high-performance computing server. The specific implementation steps are as follows: Dynamic inversion of surface combustible water content based on visual perception and standard optical calibration target assistance, such as Figure 2 As shown, to address the visual inversion distortion caused by complex lighting conditions in the field, this step constructs a dual mechanism of "hardware benchmark + algorithm correction," which is implemented as follows: Hardware Integration: A standard optical calibration target is integrated and deployed at the field monitoring station. In this embodiment, an industrial-grade 18% neutral grayscale plate is preferably used as the standard optical calibration target. It is fixed to the bottom or side of the camera pole with an extended bracket to ensure that it is always within the mechanical rotation range of the pan-tilt camera.

[0040] Data acquisition implementation: The system pulls high-definition video streams from the camera via the RTSP protocol.

[0041] Active calibration: A scheduled task is set up in the background service to drive the gimbal to the preset target coordinates every 60 minutes via control commands. An image of the target area is captured, the gain compensation coefficients of the RGB channels are calculated, and a white balance correction matrix is ​​generated.

[0042] Feature extraction and correction: After the gimbal is repositioned, video streams from the forest floor are collected, and semantic segmentation algorithms are used to extract combustible areas on the ground (such as dry grass and leaf litter). The white balance correction matrix mentioned above is applied to normalize the colors of the combustible areas on the ground, eliminating ambient light interference from dawn, dusk, and rainy weather.

[0043] Parameter inversion: The corrected image is converted to the HSV color space, and the mean hue and saturation values ​​within the region are statistically analyzed. A preset "color-moisture content mapping model" is input, and the relative moisture content of the surface combustibles is obtained by analyzing the color change from "emerald green" to "dark yellow".

[0044] Fire point location and actual slope calculation based on digital twins, such as Figure 3 As shown.

[0045] Data synchronization and fire point location: Real-time synchronization parameters: The system reads the azimuth, pitch and zoom parameters of the physical camera in real time, and calculates the vertical field of view based on the zoom parameters. These parameters are then mapped to the virtual camera components in the digital twin scene in real time to achieve virtual and real posture synchronization.

[0046] Identify the fire source: Use visual detection algorithms to locate the fire area in the video and extract the center point at the bottom of the detection box as the target pixel coordinates for localization.

[0047] Perform spatial ray projection calculation: In the digital twin scene, based on the camera parameters, a spatial ray is emitted from the virtual viewpoint toward the target pixel coordinates. The first collision point between the spatial ray and the surface of the digital elevation model (DEM) grid is detected. The coordinates of the first collision point are the three-dimensional geographic coordinates of the fire point.

[0048] Spread direction calculation: Maximum upslope vector: Obtain the surface normal at the fire point, calculate the opposite direction of the projection of the surface normal at the fire point onto the horizontal plane, which is the direction of the maximum upslope.

[0049] Vector synthesis: Real-time wind speed and direction data are acquired to construct a wind field vector. Based on the parallelogram law, the wind field vector and the maximum upslope vector are weighted and synthesized to determine the direction of fire spread under the combined influence of wind and terrain.

[0050] Actual slope calculation: Path sampling: Starting from the coordinates of the fire point, path sampling is performed in the DEM data along the direction of spread. The sampling step size is set to the grid accuracy of the digital elevation model (e.g., 30 meters) to determine the sampling endpoint.

[0051] Slope calculation: Calculate the horizontal projection distance and vertical elevation difference between the sampling start point and the end point, and use inverse trigonometric functions to solve the true slope on the dynamic path.

[0052] Real-time simulation of wildfire spread: Model calculation: The relative moisture content obtained by inversion and the actual slope obtained by solution are used as dynamic parameters and input into the forest fire spread model (Wang Zhengfei model is used in this embodiment) to calculate the forest fire spread rate; 3D visualization: The above-mentioned forest fire spread speed and spread direction are combined into a vector to construct a 3D spread vector, and the 3D spread vector is mapped onto a digital twin scene to deduce the development trend of the fire boundary and display it in 3D.

[0053] like Figure 4 and Figure 5 As shown, this embodiment adopts a modular distributed architecture based on microservices, specifically including: Multi-source sensing module: Hardware Layer: Using a field monitoring station as the integrated platform, a visible light pan-tilt camera, micro-meteorological sensors (wind speed, wind direction), and a standard optical calibration target (such as an 18% neutral grayscale plate) are uniformly installed and deployed at the station. This integrated design not only facilitates unified power supply and data transmission but also ensures that the standard optical calibration target is always within the camera's effective field of view or coverage area, providing a stable standard optical calibration reference for the white balance adaptive correction algorithm.

[0054] Streaming media distribution server: Deploys high-performance streaming media middleware, responsible for receiving RTSP video streams from cameras, transcoding and distributing them, supporting multiple concurrent streaming, and ensuring efficient transmission of video data between the intranet and the public network.

[0055] Dynamic Inversion Module (Intelligent Visual Analysis Unit): It adopts a microservice architecture and has built-in image processing algorithms. The dynamic inversion module pulls real-time video streams from the streaming media distribution server, performs semantic segmentation, adaptive white balance correction, and HSV feature extraction, and outputs the relative moisture content of surface combustibles.

[0056] The dynamic inversion module establishes a long connection with the business logic control center via WebSocket, and transmits the inverted water content parameters and fire alarm information in real time in both directions, thereby achieving millisecond-level data synchronization and command interaction.

[0057] Digital twin computing module (digital twin scenario): Developed based on real-time 3D rendering technology, it incorporates a high-precision 3D digital elevation model and constructs a digital twin scene. Specifically, the digital twin computing module includes a 3D terrain rendering unit, a spatial ray projection positioning unit, and a realistic slope calculation unit.

[0058] The digital twin computing module interacts with the business center via the HTTP protocol, receives parameters such as camera posture and wind speed, performs spatial ray projection positioning, propagation direction vector synthesis and real slope calculation, and returns the calculation results to the business center.

[0059] Business Logic Control Center (Backend Service): Built upon an enterprise-level application development framework, this system serves as the core hub of the real-time wildfire spread simulation system based on digital twins and dynamic parameter inversion, integrating a multi-source data aggregation module, a task scheduling and management module, and an alarm message distribution module. It is responsible for aggregating multi-source data (visual parameters, meteorological data, and manual instructions) and scheduling the collaborative work of the dynamic inversion module and the digital twin computing module.

[0060] The system integrates a video encoding and decoding module to extract frames, transcode, and slice key alarm video segments. It stores structured data in a relational database and video files on the local disk, achieving persistent storage of data and files.

[0061] Inference visualization module (interactive terminal): Built on a front-end development framework, it maintains a long-term connection with the back-end via WebSocket, inputting the relative moisture content and actual slope into the forest fire spread model in real time to calculate the spread rate. Combined with the spread direction, it dynamically renders the fire boundary in a digital twin scene, intuitively displaying the current spread front and the predicted future trend of the fire.

[0062] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A real-time wildfire spread prediction system based on digital twin and dynamic parameter inversion, characterized in that, It includes a multi-source sensing module, a dynamic inversion module, a digital twin computing module, and a simulation visualization module. The multi-source sensing module is integrated with a standard optical calibration target on a field monitoring station to collect visible light video streams, gimbal attitude data, and micro-meteorological data. The multi-source sensing module outputs the collected raw data to the dynamic inversion module and the digital twin computing module, respectively. The dynamic inversion module establishes an active calibration mechanism, acquires standard optical calibration target images and white balance correction matrices, performs color restoration through the white balance correction matrix, and outputs the relative moisture content of surface combustibles based on the mean hue and mean saturation in the HSV color space through a preset color-moisture content mapping model. The digital twin computing module has a built-in three-dimensional digital elevation model and constructs a digital twin scene, completing spatial ray projection positioning, propagation direction vector synthesis and real slope calculation; The outputs of the dynamic inversion module and the digital twin computing module are both transmitted as input data to the simulation visualization module, providing core input parameters for the simulation visualization module. The simulation visualization module calculates the wildfire spread rate based on the relative moisture content and the actual slope, and dynamically renders the fire boundary on the three-dimensional map in combination with the spread direction.

2. A method for real-time wildfire spread prediction using digital twins and dynamic parameter inversion, based on the real-time wildfire spread prediction system using digital twins and dynamic parameter inversion as described in claim 1, characterized in that, By setting constant optical references at field monitoring stations to eliminate ambient light interference, utilizing digital twin scenes to improve slope coordinate accuracy, determining the spread direction based on vector synthesis, and performing path sampling with the accuracy of the digital elevation model as the step size to calculate the true slope, the input error of Wang Zhengfei's forest fire spread model is significantly improved. The specific steps include: Step 1. Based on the white balance adaptive correction algorithm, perform color correction on the monitoring video stream acquired by the physical camera and the image of the preset standard optical calibration target, extract the optical features of the surface combustible area and invert the relative water content; Step 2. Synchronously map the real-time attitude parameters and zoom parameters of the physical camera to the virtual camera in the digital twin scene, so that the virtual camera and the physical camera maintain a completely consistent virtual and real attitude; emit a spatial ray from the viewpoint of the virtual camera to the pixel coordinates of the fire point target identified in the video image, and use the spatial ray projection algorithm to detect the first collision point between the emitted spatial ray and the three-dimensional digital elevation model mesh built into the digital twin scene, complete the calculation of the three-dimensional geographic coordinates of the fire point, obtain the ground projection of the surface normal vector at the corresponding location of the fire point to obtain the maximum upslope vector of the terrain, and then combine it with the wind field vector collected in real time by the wind direction and wind force sensors, and calculate the spread direction of the wildfire through vector weighted synthesis; Step 3. Starting from the fire point, and using the grid accuracy of the digital elevation model as the step size, take points forward along the spread direction obtained in Step 2, and calculate the true slope of the dynamic path. Step 4. Using the relative moisture content obtained in Step 1 and the actual slope obtained in Step 3 as dynamic parameters, input the preset Wang Zhengfei forest fire spread model to calculate the forest fire spread rate. Combine the forest fire spread rate with the spread direction determined in Step 2 to deduce the development trend of the fire boundary.

3. The method for real-time wildfire spread simulation based on digital twins and dynamic parameter inversion according to claim 2, characterized in that, The formula for calculating the spread rate in Wang Zhengfei's forest fire spread model is as follows: in, For the speed of spread, The initial spread rate, This is the moisture content correction factor obtained from the inversion. This is the wind speed correction factor. This is the corrected coefficient for the actual slope obtained from the solution.

4. The method for real-time wildfire spread prediction using digital twins and dynamic parameter inversion according to claim 2, characterized in that, Step 1, inverting the relative moisture content, includes the following steps: Step 1.1 Preset a standard optical calibration target with constant optical properties within the field of view of the physical camera or on the pan-tilt support structure, establish an active calibration mechanism, control the pan-tilt unit where the camera is located to rotate at regular intervals, and the physical camera periodically aligns with the standard optical calibration target to acquire image data of the standard optical calibration target, calculate the RGB gain deviation under the current lighting environment to generate a white balance correction matrix; and correct the white balance correction matrix based on the acquired image data. Step 1.2 Use semantic segmentation algorithm to extract the area of ​​combustible material on the ground in the surveillance video stream, and apply the white balance correction matrix obtained in Step 1.1 to perform normalized color restoration; Step 1.3 Calculate the mean hue and mean saturation of the corrected surface combustible area in the HSV color space, and output the relative moisture content of the surface combustible through the preset color-moisture content mapping model.

5. The method for real-time wildfire spread simulation based on digital twins and dynamic parameter inversion according to claim 2, characterized in that, Step 2, calculating the three-dimensional geographic coordinates of the fire point, specifically includes the following steps: Step 2.1 Read the azimuth, pitch and zoom parameters of the camera in real time, convert the zoom parameters into the vertical field of view, and map the camera parameters to the virtual camera in the digital twin scene in real time to achieve virtual and real posture synchronization. Step 2.2 Use visual detection algorithms to locate the fire area in the surveillance video stream, draw a detection box based on the identified target, and extract the center point at the bottom of the detection box as the target pixel coordinates; Step 2.3: Launch a spatial ray from the virtual camera viewpoint toward the target pixel coordinates, and detect the first collision point between the spatial ray and the digital elevation model grid. The coordinates of the first collision point are the three-dimensional geographic coordinates of the fire point.

6. The method for real-time wildfire spread prediction using digital twins and dynamic parameter inversion according to claim 2, characterized in that, In step 2, the calculation of the wildfire spread direction follows the following vector composition formula: in, The propagation direction vector, The wind field vector, It is the vector in the opposite direction of the projection of the surface normal onto the horizontal plane, i.e., the vector of the maximum upslope of the terrain; and These are the weighting coefficients.

7. The method for real-time wildfire spread simulation based on digital twins and dynamic parameter inversion according to claim 2, characterized in that, Step 3, calculating the actual physical slope, includes the following steps: Step 3.1 Using the three-dimensional geographic coordinates of the fire point as the sampling starting point, determine the sampling endpoint by moving forward along the spread direction vector with the grid accuracy of the digital elevation model as the step size; Step 3.2 Calculate the horizontal projection distance and vertical elevation difference between the sampling start point and the sampling end point, and use the inverse trigonometric function to solve the true slope of the road segment between the sampling start point and the sampling end point.

Images