Intelligent forest fire prevention method and system

Abstract

The invention discloses an intelligent forest fire prevention method and system, and aims at solving the problems of monitoring failure, information updating lag and low prediction precision of a traditional method in a severe environment by integrating multi-source real-time data and adopting deep learning and a dynamic probability model to realize accurate identification of a fire disaster and fire spreading prediction. Through an intelligent path planning algorithm and an unmanned aerial vehicle autonomous fire extinguishing technology, real-time response to dynamic changes of a fire scene and flexible adjustment of a fire extinguishing strategy are realized, the fire extinguishing efficiency and safety are improved, a fire extinguishing effect data acquisition and feedback mechanism is established, and the defect that a traditional system lacks effect evaluation after fire extinguishing is overcome. And a reliable basis is provided for subsequent model optimization and decision improvement, so that the forest fire early warning and emergency response capabilities are integrally improved.

Description

Technical Field

[0001] The present invention relates to the technical field of forest fire fighting, and specifically, to a forest intelligent fire prevention method and system. Background Art

[0002] Forest fires are one of the important disasters faced by the global ecosystem. The occurrence of fires not only destroys forest resources, but may also cause serious environmental problems and economic losses. In order to reduce the harm of forest fires, currently, methods such as manual inspections, ground monitoring equipment, and satellite remote sensing are mainly used for fire monitoring. However, these technical means still have many deficiencies in practical applications, affecting the early detection and effective control of forest fires.

[0003] Firstly, existing monitoring means often rely on a single data source. For example, some technologies only rely on visible light or infrared detection for fire recognition, and these methods may fail in environments such as bad weather and complex terrain, resulting in the failure to capture fires in a timely manner. At the same time, the coverage of some monitoring systems is limited and it is difficult to meet the requirements of large-scale and continuous monitoring.

[0004] Secondly, after a forest fire occurs, existing technologies have lags in information processing and response. For example, some technologies rely on fixed models or manual analysis for fire assessment, lacking the ability of dynamic adjustment, resulting in low accuracy of prediction results. In addition, due to the limitation of data processing capabilities, some systems are difficult to quickly integrate various environmental information, thus affecting the efficiency of emergency decision-making.

[0005] In addition, during the fire response process, traditional operation methods mainly rely on manual decision-making or preset fixed plans, lacking the ability to adapt to the dynamic changes of the fire scene in real time. For example, during the spread of the fire, factors such as wind direction and terrain may change, and existing technologies are difficult to adjust the fire extinguishing strategy in a timely manner, resulting in a decrease in operation efficiency and even affecting the fire extinguishing effect. In addition, after the fire is disposed of, existing systems often lack a complete data collection and analysis mechanism, unable to effectively evaluate the actual effect of fire extinguishing measures, and also difficult to provide reliable support for subsequent optimization. Summary of the Invention

[0006] The purpose of the present invention is to provide a forest intelligent fire prevention method and system, which can at least solve the problems proposed in the above background art.

[0007] According to one aspect of the present application, a forest intelligent fire prevention method is provided, including the following steps:

[0008] The inspection unmanned aerial vehicle group performs fire source detection and real-time collection of forest fire images, and transmits the collected image data back to the cloud platform;

[0009] The cloud platform extracts feature vectors from the processed image data through the Convlstm deep learning architecture and the GWO algorithm, determines the fire recognition model, and judges the location and scale of the fire occurrence;

[0010] The cloud platform constructs a multi-dimensional dynamic fire spread model and predicts the fire spread path based on the GFS-MD model by integrating fire data, meteorological data, vegetation distribution data, and terrain elevation data;

[0011] The cloud platform uses the butterfly optimization algorithm to determine the optimal waypoint sequence path of the unmanned fire extinguishing unit based on the fire spread path, the terrain elevation data, and the real-time meteorological data, and sends it to the unmanned fire extinguishing unit;

[0012] The unmanned fire extinguishing unit flies according to the optimal waypoint sequence path, and combines the gravitational acceleration and the wind direction compensation algorithm to perform U-shaped point-to-point bombing on the core area of the fire source.

[0013] Preferably, extracting feature vectors from the processed image data through the Convlstm deep learning architecture and the GWO algorithm, and determining the fire recognition model includes:

[0014] Preprocess the forest fire image data collected by the patrol unmanned unit in real time, including denoising, image enhancement, and normalization, to obtain the preprocessed forest fire image data;

[0015] Through the convolutional neural network in the Convlstm deep learning architecture, extract the spatial features of the fire in the preprocessed forest fire image data, including the shape, size, color distribution of the flame, and the core area of the fire source;

[0016] Through the long short-term memory network in the Convlstm deep learning architecture, capture the dynamic change process of the fire, including the spread trend of the flame, the change of the fire intensity, and the temporal characteristics of the fire development;

[0017] Fuse the extracted spatial features and time series features to generate the feature vector of the fire;

[0018] Use the gray wolf optimization algorithm to optimize the feature vector of the fire:

[0019]

[0020] Among them, X i represents the i-th feature vector, X optimal represents the optimal feature vector, and w i is the weight coefficient;

[0021] Based on the optimized fire feature vector, train the fire recognition model;

[0022] Based on the trained fire recognition model, analyze the real-time collected forest fire image data to determine the location and scale of the fire.

[0023] Preferably, the cloud platform constructs a multi-dimensional dynamic fire spread model and predicts the fire spread path, which specifically includes the following steps:

[0024] Establish a fire state set S s ={S1, S2,..., S n}), where each state S i represents a different fire intensity level. Based on the fire ground temperature, flame height, wind speed and direction, and combustible content, construct a state space through a Markov model, where

[0025] S=(T level , H level , V level , θ dir , C level )

[0026] T level represents the flame temperature level, H level represents the flame height level, V level represents the wind speed level, θ dir represents the wind direction, and C level represents the combustible content level.

[0027] Preferably, according to the historical fire data, statistically analyze the transition frequency of the state space, and determine the state transition probability through maximum likelihood estimation as:

[0028] P ij =P(S t =S j |S t-1 =S i )

[0029] where,

[0030] P ij represents the probability that the fire spreads from state S i to state S j .

[0031] S t represents the state of the fire at time t,

[0032] S j represents the possible state that the fire may transfer to at the next time point, the possible state target during the fire development process,

[0033] The state transition probability matrix P represents the probability of the fire transferring from one state to another, and is used to describe the dynamic process of the fire evolving over time;

[0034] Update the state transition probability according to the current fire situation data in combination with the Bayesian inference method:

[0035]

[0036] where X t represents the real-time forest fire situation data and the meteorological data,

[0037] P(S t |X t ) is the posterior probability that the fire situation is in S t under the condition of the real-time forest fire situation data and the meteorological data X t ,

[0038] P(X t |S t ) represents the likelihood probability that X t appears when the fire situation is in S t ,

[0039] P(X t ) represents the evidence probability of observing X t ,

[0040] P(S t ) represents the prior probability when the fire situation is in state S t ;

[0041] Combine time series analysis and generate the fire field boundary and the fire spread direction to predict the fire spread path.

[0042] Preferably, use the butterfly algorithm to calculate the optimal waypoint sequence path of the UAV based on the fire field boundary, the terrain elevation model and the wind speed and direction data,

[0043]

[0044] where

[0045] J represents the total optimization objective function, which is used to evaluate the quality of the waypoint sequence path;

[0046] n represents the total number of waypoints;

[0047] (x i , y i ) represents the coordinate position of the i-th waypoint;

[0048] θ i represents the heading angle of the i-th path;

[0049] w1, w2, and w3 respectively represent the weight coefficients of the fire field boundary distance, the terrain elevation, and the wind speed and direction;

[0050] f dist (x i ,y i ) represents the function of the distance from the fire boundary;

[0051] f elev (x i ,y i ) represents the terrain elevation function;

[0052] f wind (x i ,y i ,θ i ) represents the function of the influence of wind speed and direction.

[0053] Preferably, the formation of fire-fighting drones flies according to the optimal waypoint sequence. Based on the gravity acceleration and wind direction compensation algorithm, the formula for the U-shaped point-to-point bombing in the core area of the fire source is:

[0054]

[0055] Among them,

[0056] Q is the delivery amount of fire-extinguishing bombs, ΔT is the temperature difference between the target area of the fire field and the surrounding environment, is the rate of change of the temperature difference, W is the influence factor of wind speed, H is the influence factor of terrain elevation, k p , α, k w , k h are the weight coefficients of temperature difference, rate of change of temperature difference, wind speed and terrain elevation respectively.

[0057] According to another aspect of the present invention, a forest intelligent fire prevention system is provided, including:

[0058] Patrol drone group: The patrol drone group performs fire source detection and real-time collection of forest fire images, and transmits the collected image data back to the cloud platform;

[0059] Cloud platform: The cloud platform extracts feature vectors from the processed image data through the Convlstm deep learning architecture and the GWO algorithm, determines the fire recognition model, and judges the location and scale of the fire; Based on the GFS-MD model, it fuses fire data, meteorological data, vegetation distribution data and terrain elevation data to construct a multi-dimensional dynamic fire spread model and predict the fire spread path; It adopts the butterfly optimization algorithm, and based on the fire spread path, the terrain elevation data and the real-time meteorological data, determines the optimal waypoint sequence path of the fire-fighting drone group and sends it to the fire-fighting drone group;

[0060] Fire extinguishing unmanned aerial vehicle group: The fire extinguishing unmanned aerial vehicle group flies along the optimal waypoint sequence path, combines the gravitational acceleration and wind direction compensation algorithm, and performs U-shaped point-to-point bombing on the core area of the fire source.

[0061] According to another aspect of the present application, there is provided an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the program to implement the forest intelligent fire prevention method provided by the present application.

[0062] According to another aspect of the present application, there is provided a computer-readable storage medium, on which a computer program is stored, and the program is executed by a processor to implement the forest intelligent fire prevention method provided by the present application.

[0063] According to another aspect of the present application, there is provided a computer program product, and the computer program is executed to implement the forest intelligent fire prevention method provided by the present application.

[0064] Compared with the prior art, the beneficial effects of the forest intelligent fire prevention method and system provided by the present application are as follows: By integrating multi-source real-time data, deep learning and dynamic probability models are used to achieve accurate identification of fires and prediction of fire spread, solving the problems of monitoring failure, information update lag, and low prediction accuracy of traditional methods in harsh environments. At the same time, through the intelligent path planning algorithm and unmanned aerial vehicle autonomous fire extinguishing technology, real-time response to the dynamic changes of the fire scene and flexible adjustment of the fire extinguishing strategy are realized, improving the fire extinguishing efficiency and safety, and establishing a data collection and feedback mechanism for the fire extinguishing effect, making up for the lack of post-fire effect evaluation in traditional systems, providing a reliable basis for subsequent model optimization and decision improvement, and thus overall enhancing the forest fire early warning and emergency response capabilities. Description of the Drawings

[0065] Figure 1 It is a flowchart of the forest intelligent fire prevention method according to an embodiment of the present application. Detailed Embodiments

[0066] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0067] Please refer to Figure 1 , according to an embodiment of the present invention, there is provided a forest intelligent fire prevention method, including the following steps:

[0068] S10: The patrol UAV group performs fire source detection and collects forest fire images in real time, and transmits the collected image data back to the cloud platform.

[0069] The patrol UAV group performs fire source detection tasks in the forest area, uses on-board camera equipment to collect forest fire image data in real time, and transmits the collected real-time forest fire image data back to the cloud platform through a wireless communication network for subsequent fire recognition and analysis.

[0070] S20: The cloud platform extracts feature vectors from the processed image data through the Convlstm (Convolutional Long Short-Term Memory) deep learning architecture and the GWO algorithm, determines the fire recognition model, and judges the location and scale of the fire.

[0071] After receiving the real-time forest fire image data transmitted back by the patrol UAV group, the cloud platform processes the image data based on the Convlstm deep learning architecture. The Convlstm model extracts the spatial features in the image and combines the time series data to identify the dynamic changes of the fire situation.

[0072] Subsequently, the cloud platform applies the GWO algorithm to optimize the extracted feature vectors to improve the accuracy of the fire recognition model. Based on the optimized model (here, is the optimized model the fire recognition model?), the cloud platform judges the location and scale of the fire and forms a fire situation analysis result to determine the location and scale of the fire.

[0073] S30: The cloud platform constructs a multi-dimensional dynamic fire spread model and predicts the fire spread path by fusing fire data, meteorological data, vegetation distribution data, and terrain elevation data based on the GFS-MD (Global Fire Spread-Multidimensional Dynamics) model.

[0074] The cloud platform first calls the fire situation analysis result from the fire perception module, and then combines the real-time meteorological data obtained from the meteorological monitoring station, the vegetation distribution data determined by GIS remote sensing technology and combined with NDVI, and the terrain elevation model data obtained from the cloud database.

[0075] On this basis, the cloud platform uses the GFS-MD model to simulate and analyze the fire spread process, calculates the dynamic change trends of the fire spread direction, the fire field boundary, and the temperature distribution, predicts the fire spread path. The model considers various environmental impact factors inside and outside the fire field during the simulation process, enabling the prediction results to reflect the gradual expansion of the fire field boundary and the changes in the fire situation. Finally, the cloud platform generates a sequence of predicted infrared images to visually display the future temperature distribution of the fire field, providing a basis for subsequent fire extinguishing tasks.

[0076] S40: The cloud platform adopts the butterfly optimization algorithm. Based on the fire spread path, terrain elevation data, and real-time meteorological data, it determines the optimal waypoint sequence path for the unmanned fire extinguishing aircraft group and sends it to the unmanned fire extinguishing aircraft group.

[0077] During the optimization process, the butterfly optimization algorithm constructs an objective function to balance the flight time, energy consumption, and safety obstacle avoidance requirements, and adjusts the waypoints according to the dynamic changes of the fire field. Finally, the calculated optimal waypoint sequence is stored in the cloud in the format of a standard wireless communication protocol and sent to the unmanned fire extinguishing aircraft group through the wireless communication network, providing real-time and dynamically updated flight path planning instructions for the unmanned aircraft.

[0078] S50: The unmanned fire extinguishing aircraft group flies according to the optimal waypoint sequence path, and combines the gravitational acceleration and wind direction compensation algorithm to perform U-shaped point-to-point bombing on the core area of the fire source.

[0079] The unmanned fire extinguishing aircraft group performs the fire extinguishing task according to the optimal waypoint sequence path sent by the cloud platform. During the flight process, the unmanned fire extinguishing aircraft group combines the gravitational acceleration and wind direction compensation algorithm to accurately adjust the dropping position and density of the fire extinguishing bombs to ensure the fire extinguishing effect.

[0080] Meanwhile, the unmanned fire extinguishing aircraft group monitors the changes in the fire field in real time and transmits the fire extinguishing effect data back to the cloud platform for the subsequent optimization of the fire spread prediction model and the adjustment of the fire extinguishing strategy.

[0081] In one embodiment, feature vectors are extracted from the processed image data through the Convlstm deep learning architecture and the GWO algorithm, and the fire recognition model is determined to include:

[0082] S1110: Preprocess the forest fire image data collected by the patrol unmanned aircraft group in real time, including denoising, image enhancement, and normalization, to obtain the preprocessed forest fire image data.

[0083] In forest areas with high fire risks, unmanned aircraft regularly or continuously perform cruise tasks and use the equipped high-definition and thermal infrared cameras to collect forest fire images. After these image data are preprocessed (including denoising, image enhancement, and normalization), they are uploaded to the cloud platform in real time through the wireless communication network to ensure that the latest fire information can be quickly summarized.

[0084] S1120: Extract the spatial features of the fire in the pre-processed forest fire image data through the convolutional neural network in the Convlstm deep learning architecture, including the shape, size, color distribution of the flame, and the core area of the fire source.

[0085] After receiving the image data, the cloud platform uses the Convlstm deep learning architecture to extract the spatial and temporal features in the image. Through the convolutional neural network part, the cloud platform captures the spatial features such as the shape, size, and color distribution of the fire area; at the same time, the long short-term memory network part can capture the dynamic changes during the fire development process. Combining the feature vectors optimized by the grey wolf optimization algorithm, the cloud platform constructs a fire recognition model to judge the location and scale of the fire occurrence in real time, providing accurate basic information for the next step of fire trend prediction.

[0086] S1130: Capture the dynamic change process of the fire through the long short-term memory network in the Convlstm deep learning architecture, including the spreading trend of the flame, the change of the fire intensity, and the temporal characteristics of the fire development.

[0087] After a fire occurs, the cloud platform constructs a multi-dimensional dynamic fire spread model based on the GFS-MD model by integrating the fire situation analysis results, real-time meteorological data, vegetation distribution data, and terrain elevation data. By simulating the fire spread process, the model predicts the future fire spread direction, fire field boundary, and temperature distribution, and at the same time generates a predicted infrared image sequence to visually display the future state of the fire field. These prediction results can help the command center deploy resources in advance and formulate emergency measures.

[0088] S1140: Integrate the extracted spatial features and time series features to generate the feature vectors of the fire, and train the fire recognition model based on the optimized fire feature vectors.

[0089] Using the fire field boundary data output by the fire trend prediction module, the terrain elevation data obtained from the cloud GIS database, and the real-time wind speed and wind direction data from the meteorological monitoring station, the cloud platform uses an improved butterfly optimization algorithm to calculate the optimal waypoint sequence path of the fire extinguishing unmanned aerial vehicle group. This algorithm comprehensively considers the flight time, energy consumption, and safety obstacle avoidance requirements to ensure that the UAV path can avoid the dangerous areas and terrain obstacles of the fire field while effectively reducing the energy consumption. After the calculation is completed, the optimal waypoint sequence is sent to the fire extinguishing unmanned aerial vehicle group through wireless communication.

[0090] Among them, the grey wolf optimization algorithm is used to optimize the feature vectors of the fire:

[0091]

[0092] Among them, X i represents the i-th feature vector, and X optimal represents the optimal feature vector, and wi is the weight coefficient.

[0093] S1150: Based on the trained fire recognition model, analyze the real-time collected forest fire image data to determine the location and scale of the fire.

[0094] The unmanned fire extinguishing aircraft group executes the fire extinguishing task according to the waypoint sequence issued by the cloud platform. During the flight, the unmanned aircraft adjusts the dropping density of the fire extinguishing bombs through the gravity acceleration and wind direction compensation algorithms according to the temperature data collected in real time by the onboard infrared thermal imaging device, ensuring that the fire extinguishing bombs accurately cover the core area of the fire source. At the same time, the unmanned aircraft collects the fire extinguishing effect data and feeds it back to the cloud platform in real time for the system to further optimize the fire spread prediction and path planning strategies, realizing closed-loop adaptive control.

[0095] In one embodiment, specifically, the cloud platform constructs a multi-dimensional dynamic fire spread model and predicts the fire spread path, which specifically includes the following steps:

[0096] S1210: Establish a fire state set, where each state represents a different fire intensity level. Based on the fire field temperature, flame height, wind speed and direction, and combustible content, construct a state space through a Markov model, where

[0097] S = (T level , H level , V level , θ dir , C level )

[0098] T level represents the flame temperature level, H level represents the flame height level, V level represents the wind speed level, θ dir represents the wind direction, and C level represents the combustible content level.

[0099] First, the system constructs a fire state space according to the actual monitored data of the fire field temperature, flame height, wind speed and direction, and combustible content. Each state represents a level of fire intensity, such as the initial, stable, intense, and declining states, and is transmitted to the cloud platform after preprocessing for establishing the state set.

[0100] S1220: According to the historical fire data, count the transition frequencies between states, and determine the state transition probabilities through maximum likelihood estimation.

[0101] Subsequently, the cloud platform uses historical fire data to statistically analyze the transitions between various fire states. By calculating the number of transitions between different fire states and using the maximum likelihood estimation method to determine the state transition probability matrix, it can obtain the probability distribution of the fire transitioning from one stage to another, providing basic data for predicting the fire dynamics.

[0102] S1230: Combine the Bayesian inference method to update the state transition probability according to the current fire situation data.

[0103] During the real-time occurrence of a fire, the cloud platform combines the currently monitored fire situation data to update the original state transition probability. Using the Bayesian inference method, the cloud platform corrects the prior probability of the fire state based on the latest observation data, and then obtains a posterior probability distribution that better conforms to the current fire scene situation, which can reflect the fire trend in real time and improve the prediction accuracy.

[0104] S1240: Combine time series analysis, generate the fire scene boundary and the fire spread direction, and predict the fire spread path.

[0105] Finally, the cloud platform uses the updated state transition probability and combines the time series analysis method to predict the fire scene boundary and the fire spread direction. By modeling the evolution of the fire scene boundary over time, the system can predict the fire spread path and the fire scene expansion trend, and the prediction results are used for subsequent resource scheduling and emergency plan formulation.

[0106] Among them, the determination of the state transition probability is:

[0107] P ij =P(S t =S j |S t-1 =S i )

[0108] Among them,

[0109] P ij represents the probability that the fire transitions from state S i to state S j .

[0110] S t represents the state of the fire at time t,

[0111] S j represents the possible state that the fire may transition to at the next time point, the possible state target during the fire development process.

[0112] The state transition probability matrix P represents the probability of the fire transitioning from one state to another, and is used to describe the dynamic process of the fire evolving over time;

[0113] Update the state transition probability according to the current fire situation data in combination with the Bayesian inference method:

[0114]

[0115] Among them, X t represents the real-time forest fire situation data and meteorological data,

[0116] P(S t |X t ) represents the posterior probability that the fire situation is in S t under the condition of the real-time forest fire situation data and meteorological data X t ;

[0117] P(X t |S t ) represents the likelihood probability that X t is observed when the fire situation is in S t ;

[0118] P(X t ) represents the evidence probability of observing X t ;

[0119] P(S t ) represents the prior probability when the fire situation is in state S t ;

[0120] Combine time series analysis, and generate the fire field boundary, the fire spread direction, and predict the fire spread path.

[0121] In one embodiment, specifically, the butterfly algorithm is used to calculate the optimal waypoint sequence path of the UAV based on the fire field boundary, the terrain elevation model, and the wind speed and direction data,

[0122]

[0123] Among them,

[0124] J represents the total optimization objective function, which is used to evaluate the quality of the waypoint sequence path;

[0125] n represents the total number of waypoints;

[0126] (x i , y i ) represents the coordinate position of the i-th waypoint;

[0127] θ i represents the heading angle of the i-th path;

[0128] w1, w2, and w3 respectively represent the weight coefficients of the fire field boundary distance, the terrain elevation, and the wind speed and direction;

[0129] f dist(x i ,y i ) represents the fire field boundary distance function;

[0130] f elev (x i ,y i ) represents the terrain elevation function;

[0131] f wind (x i ,y i ,θ i ) represents the wind speed and direction influence function.

[0132] In one embodiment, the fire extinguishing drone formation flies according to the optimal waypoint sequence. Based on the gravitational acceleration and wind direction compensation algorithm, the formula for the U-shaped point-to-point bombing execution in the fire source core area is:

[0133]

[0134] Among them,

[0135] Q is the delivery amount of fire extinguishing bombs, ΔT is the temperature difference between the fire field target area and the surrounding environment, is the rate of change of the temperature difference, W is the influence factor of wind speed, H is the influence factor of terrain elevation, k p , α, k w , k h are the weight coefficients of temperature difference, rate of change of temperature difference, wind speed, and terrain elevation respectively.

[0136] According to another aspect of the present invention, a forest intelligent fire prevention system includes:

[0137] Patrol drone group: The patrol drone group performs fire source detection and real-time collection of forest fire images, and transmits the collected image data back to the cloud platform;

[0138] Cloud platform: The cloud platform extracts feature vectors from the processed image data through the Convlstm deep learning architecture and the GWO algorithm, determines the fire recognition model, and judges the location and scale of the fire;

[0139] The cloud platform constructs a multi-dimensional dynamic fire spread model and predicts the fire spread path based on the GFS-MD model by fusing fire data, meteorological data, vegetation distribution data, and terrain elevation data;

[0140] The cloud platform uses the butterfly optimization algorithm to determine the optimal waypoint sequence path of the fire extinguishing drone group based on the fire spread path, terrain elevation data, and real-time meteorological data, and sends it to the fire extinguishing drone group;

[0141] Fire-fighting unmanned aerial vehicle group: The fire-fighting unmanned aerial vehicle group flies according to the optimal waypoint sequence path, combines the gravitational acceleration and the wind direction compensation algorithm, and performs U-shaped point-to-point bombing on the core area of the fire source.

[0142] In summary, the forest intelligent fire prevention system integrates the patrol unmanned aerial vehicle, the cloud platform and the fire-fighting unmanned aerial vehicle group to realize the full-process closed-loop control of fire detection, identification, prediction, path planning and precise fire extinguishing. The patrol unmanned aerial vehicle regularly cruises in the fire risk area, collects high-resolution fire images in real time, and quickly transmits them back to the cloud platform. The cloud platform uses the Convlstm deep learning architecture and the GWO algorithm to extract key features from the image data and accurately judge the location and scale of the fire; at the same time, based on the GFS-MD model, it fuses multi-source data from sensors, meteorological monitoring stations, GIS remote sensing, etc. to construct a multi-dimensional dynamic fire spread model to predict the fire spread path and the fire field boundary.

[0143] Furthermore, the cloud platform uses an improved butterfly optimization algorithm to calculate the optimal unmanned aerial vehicle waypoint sequence path and sends it to the fire-fighting unmanned aerial vehicle group through wireless communication. The fire-fighting unmanned aerial vehicle group flies according to the planned path, combines the gravitational acceleration and the wind direction compensation algorithm, and performs U-shaped point-to-point bombing on the core area of the fire source to ensure the accuracy and efficiency of the fire extinguishing operation. This system improves the fire warning response speed and the fire extinguishing accuracy, providing reliable technical support for forest fire prevention and control.

[0144] Parts not involved in the present invention are the same as or can be implemented using the prior art. Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A forest intelligent fire prevention method, characterized in that: The following steps are involved: The patrol drone group performs fire source detection and collects forest fire images in real time, and transmits the collected image data back to the cloud platform; The cloud platform uses the ConvLSTM deep learning architecture and GWO algorithm to extract feature vectors from the processed image data, determine the fire recognition model, and determine the location and scale of the fire; The cloud platform integrates fire data, meteorological data, vegetation distribution data, and terrain elevation data based on the GFS-MD model to build a multi-dimensional dynamic fire spread model and predict the fire spread path; The cloud platform uses a butterfly optimization algorithm to determine the optimal waypoint sequence path of the fire-fighting drone group based on the fire spread path, the terrain elevation data and real-time meteorological data, and sends it to the fire-fighting drone group; The fire-fighting drone group flies according to the optimal waypoint sequence path, combines gravity acceleration and wind direction compensation algorithm, and performs U-shaped point-to-point bombing on the core area of ​​the fire source.

2. The intelligent forest fire prevention method according to claim 1, characterized in that: The ConvLSTM deep learning architecture and GWO algorithm are used to extract feature vectors from the processed image data to determine the fire recognition model, including: Preprocess the forest fire image data collected in real time by the patrol drone group, including denoising, image enhancement and normalization, to obtain preprocessed forest fire image data; Through the convolutional neural network in the ConvLSTM deep learning architecture, the spatial characteristics of the fire in the preprocessed forest fire image data are extracted, including the shape, size, color distribution of the flame and the core area of ​​the fire source; Through the long short-term memory network in the ConvLSTM deep learning architecture, the dynamic changes of the fire are captured, including the spread trend of the flame, the change of the fire intensity, and the temporal characteristics of the fire development; The extracted spatial features and time series features are integrated to generate the feature vector of the fire; Use the Gray Wolf optimization algorithm to optimize the fire feature vector: Among them, X i represents the i-th eigenvector, X optimal represents the optimal eigenvector, w i is the weight coefficient; Based on the optimized fire feature vector, a fire recognition model is trained; Based on the trained fire recognition model, the real-time collected forest fire image data is analyzed to determine the location and scale of the fire.

3. The intelligent forest fire prevention method according to claim 1, characterized in that: The cloud platform builds a multi-dimensional dynamic fire spread model and predicts the fire spread path, which includes the following steps: Establish fire state set S s ={S1,S2,...,S n }, where each state S i Representing different levels of fire intensity, the state space is constructed through the Markov model based on fire temperature, flame height, wind speed and direction, and combustible content. S=(T level ,H level ,V level ,θ dir ,C level ) T level Indicates the flame temperature level, H level Indicates the flame height level, V level Indicates wind speed level, θ dir Indicates wind direction, C level Indicates the combustible content level.

4. The intelligent forest fire prevention method according to claim 3, characterized in that: The transition frequency of the state space is statistically calculated based on the historical fire data, and the state transition probability is determined by maximum likelihood estimation: P ij =P(S t =S j |S t-1 =S i ) in, P ij Indicates that the fire changes from state S i Transfer to state S j The probability of S t represents the state of the fire at time t, S j Indicates the state that the fire may transfer to at the next time point, and the possible state target during the development of the fire. The state transition probability matrix P represents the probability of a fire transitioning from one state to another, and is used to describe the dynamic process of fire evolution over time; Combined with the Bayesian inference method, the state transition probability is updated according to the current fire data: Among them, X t represents the real-time forest fire data and the meteorological data, P(S t |X t ) indicates that in providing real-time forest fire data and the meteorological data X t In the case of fire at S t The posterior probability of P(X t |S t ) represents the fire at S t When X t The likelihood of occurrence, P(X t ) represents the observation X t The probability of evidence, P(S t ) represents the fire is in state S t The prior probability when ; Combined with time series analysis, the fire scene boundary and the fire spread direction are generated to predict the fire spread path.

5. The intelligent forest fire prevention method according to claim 1, characterized in that: The steps of calculating the optimal waypoint sequence path of the UAV using the butterfly algorithm based on the fire scene boundary, the terrain elevation model and the wind speed and direction data are as follows: in, J represents the overall optimization objective function, which is used to evaluate the quality of the waypoint sequence path; n represents the total number of waypoints; (x i ,y i ) represents the coordinate position of the i-th waypoint; θ i represents the heading angle of the i-th path; w1, w2, and w3 represent the weight coefficients of fire boundary distance, terrain elevation, and wind speed and direction, respectively; f dist (x i ,y i ) represents the fire boundary distance function; f elev (x i ,y i ) represents the terrain elevation function; f wind (x i ,y i ,θ i ) represents the wind speed and direction influence function.

6. The intelligent forest fire prevention method according to claim 1, characterized in that: in, The firefighting drone formation flies according to the optimal waypoint sequence, and based on the gravity acceleration and wind direction compensation algorithm, the steps of performing U-shaped point-to-point bombing on the core area of ​​the fire source are as follows: in, Q is the amount of fire extinguishing bombs released, ΔT is the temperature difference between the target area of ​​the fire scene and the surrounding environment, is the rate of temperature difference change, W is the influencing factor of wind speed, H is the influencing factor of terrain elevation, k p ,α,k w , k h They are the weight coefficients of temperature difference, temperature difference changing rate, wind speed and terrain elevation respectively.

7. A forest intelligent fire prevention system, characterized in that: include: Patrol drone group: The patrol drone group performs fire source detection and collects forest fire images in real time, and transmits the collected image data back to the cloud platform; Cloud platform: The cloud platform extracts feature vectors from processed image data through the Convlstm deep learning architecture and GWO algorithm, determines the fire recognition model, and judges the location and scale of the fire. Based on the GFS-MD model, it integrates fire data, meteorological data, vegetation distribution data, and terrain elevation data to build a multi-dimensional dynamic fire spread model and predict the fire spread path. The butterfly optimization algorithm is used to determine the optimal waypoint sequence path of the fire-fighting drone group based on the fire spread path, the terrain elevation data, and real-time meteorological data, and sends it to the fire-fighting drone group. Firefighting drone group: The firefighting drone group flies according to the optimal waypoint sequence path, combines gravity acceleration and wind direction compensation algorithm, and performs U-shaped point-to-point bombing on the core area of ​​the fire source.

8. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the intelligent forest fire prevention method as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the intelligent forest fire prevention method as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that The computer program is executed to implement the intelligent forest fire prevention method as described in any one of claims 1-6.

Images