Unmanned aerial vehicle cluster cooperative smoke dispatching method for maximizing smoke coverage

Abstract

The present application provides a kind of unmanned aerial vehicle cluster cooperative smoke dispatching method for maximizing smoke coverage, solve the technical problems of low smoke diffusion prediction accuracy, poor unmanned aerial vehicle cluster cooperation, weak dynamic adaptability of scheduling strategy and difficult to maximize coverage in existing method.The present application includes the following steps: step 1, collect real-time operation data and environment data of smoke unmanned aerial vehicle and pretreat;Step 2, improve ReTUNet model, define the improved ReTUNet model as IReTUNet model, predict smoke diffusion state using IReTUNet model, output smoke concentration distribution matrix, calculate smoke coverage based on the matrix and identify coverage blind area;Step 3, improve IA * algorithm, optimize its heuristic function;Through the improved IA * algorithm, the optimal scheduling strategy is output, which guides the unmanned aerial vehicle to execute the smoke task.

Description

Technical Field

[0001] This invention relates to the field of cluster collaborative scheduling, and specifically to a method for collaborative smoke generation scheduling of UAV clusters aimed at maximizing smoke screen coverage. Background Technology

[0002] In modern military camouflage, emergency rescue, and environmental governance, smoke screens are widely used as a low-cost, high-efficiency means of concealment and protection, such as blocking enemy reconnaissance, preventing the spread of toxic gases, and covering the transfer of personnel and equipment. With the rapid development of drone technology, using drone swarms to collaboratively generate smoke instead of traditional manual or single-drone smoke generation has become a core direction for improving smoke screen coverage and optimizing its effectiveness. Drone swarms have the advantages of high mobility, flexible deployment, and the ability to generate smoke simultaneously from multiple points, enabling the rapid construction of large-scale, high-density smoke screen barriers that are suitable for complex terrain and dynamic mission requirements.

[0003] However, current methods for coordinating smoke generation in UAV swarms still face numerous technical bottlenecks, making it difficult to maximize smoke screen coverage and severely limiting their application effectiveness. Firstly, the smoke screen diffusion process is highly dynamic in terms of time and space. Influenced in real-time by environmental factors such as wind speed, wind direction, and temperature, the smoke screen's diffusion trajectory, concentration distribution, and coverage area exhibit complex nonlinear characteristics. Traditional scheduling methods often use simplified physical models such as Gaussian plumes and smoke clouds for diffusion prediction. These models make stringent assumptions and cannot accurately capture the spatiotemporal correlation characteristics of smoke screen diffusion, leading to significant deviations in smoke screen coverage assessments. This, in turn, misleads UAV scheduling strategies, resulting in coverage blind spots or overlapping smoke generation. Secondly, traditional models rely on complex physical formula derivations, incurring high computational costs and being sensitive to initial conditions, making them unsuitable for real-time scheduling.

[0004] In summary, existing UAV swarm collaborative smoke generation scheduling methods suffer from low smoke spread prediction accuracy, inaccurate coverage assessment, poor UAV coordination, and weak dynamic adaptability of scheduling strategies. Therefore, there is an urgent need for a UAV swarm collaborative smoke generation scheduling method aimed at maximizing smoke coverage, thereby improving the application efficiency of UAV swarm collaborative smoke generation. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a collaborative smoke generation scheduling method for UAV swarms aimed at maximizing smoke screen coverage, combining IReTUNet and IA * The algorithm addresses the technical problems of low smoke diffusion prediction accuracy, poor drone swarm coordination, weak dynamic adaptability of scheduling strategies, and difficulty in maximizing coverage in existing methods. It achieves accurate prediction of smoke diffusion, accurate assessment of coverage, and efficient collaborative scheduling of drone swarms, thereby improving smoke coverage efficiency and adapting to the needs of smoke generation tasks in complex scenarios.

[0006] The technical solution adopted by this invention to solve the above problems is: a drone swarm collaborative smoke generation scheduling method for maximizing smoke screen coverage, comprising the following steps:

[0007] Step 100: Collect real-time operational data and environmental data of the smoke-generating drone and perform preprocessing;

[0008] Step 200: Improve the ReTUNet model by defining the improved model as the IReTUNet model. Use the IReTUNet model to predict the smoke diffusion state, output the smoke concentration distribution matrix, and calculate the smoke coverage rate and identify coverage blind spots based on this matrix. The improvements include two aspects:

[0009] 1. In the recursive Transformer encoder, a spatial decay mechanism based on Manhattan distance is introduced to replace the traditional Transformer self-attention mechanism. When calculating the attention weight between two spatial points, a Manhattan distance decay term is added.

[0010] Second, a multi-scale feature fusion module is added to the U-Net decoder to fuse high-level feature vectors of different scales output by the recursive Transformer encoder.

[0011] Step 300, for IA * The algorithm is improved by optimizing its heuristic function; the improved IA is then used. * The algorithm performs collaborative scheduling optimization for UAV clusters, completing smoke emission priority allocation, path planning, and smoke emission adjustment; finally, it outputs the optimal scheduling strategy to guide UAVs in performing smoke emission tasks.

[0012] Step 100 of the present invention includes the following steps:

[0013] Step 101: Collect real-time drone operation data: Determine the number N of drones participating in collaborative smoke generation, and collect the current position coordinates, flight speed, remaining battery power, current smoke generation amount, and smoke generation rate of each drone in real time through the GPS positioning module, attitude sensor, and smoke generation sensor carried by each drone; Collect environmental data: Collect wind speed, wind direction, and ambient temperature through ground environmental monitoring equipment;

[0014] Step 102: Standardize the real-time operation data and environmental data of the UAV;

[0015] Step 103: Convert the standardized data into a spatiotemporal feature matrix and IA adapted to the input of the IReTUNet model. * The parameter vector input to the algorithm is stored in the database of the ground control station.

[0016] In step 200 of this invention, the formula for calculating the attention weight between two spatial points is as follows:

[0017] ,

[0018] Where Attention(i,j) is the attention weight between the i-th spatial point and the j-th spatial point, Sim(i,j) is the feature similarity between the i-th spatial point and the j-th spatial point, α is the spatial decay coefficient, and K is the total number of spatial points; Sim(i,k) is the feature similarity between the i-th spatial point and the k-th spatial point, d man (i,k) is the Manhattan distance between the i-th spatial point and the k-th spatial point, d man (i,j) is the Manhattan distance between the i-th spatial point and the j-th spatial point, and its calculation formula is: d man (i,j)=|x i -x j |+|y i -y j |, where x i x j These are the x-coordinates of the i-th and j-th spatial points, respectively, and y i y j These are the ordinates of the i-th and j-th spatial points, respectively.

[0019] Step 200 of the present invention further includes the following steps:

[0020] Step (1) Model training: Train the IReTUNet model. During the training process, mean squared error is used as the loss function and Adam optimizer is used. The model parameters are updated through backpropagation until the model converges.

[0021] Step (2) Smoke diffusion prediction: Input the standardized data into the trained IReTUNet model, and the IReTUNet model performs the prediction process;

[0022] Step (3) Smoke Coverage Calculation and Coverage Blind Spot Identification: Based on the IReTUNet model, output the smoke concentration distribution matrix, combine it with the smoke concentration threshold, calculate the current smoke coverage, and identify coverage blind spots;

[0023] Step (4) outputs the current smoke cover coverage, smoke concentration distribution matrix, and coverage blind zone coordinate set to the ground control station in real time as IA * The core input for the algorithm to perform cooperative scheduling optimization.

[0024] The prediction process of step (2) of step 200 of the present invention includes the following steps:

[0025] (1) The input layer of the IReTUNet model converts the standardized data into a spatiotemporal feature matrix, and the number of features is the number of categories of the input data;

[0026] (2) The recursive Transformer encoder processes the spatiotemporal feature matrix, captures the temporal correlation of smoke diffusion through a bidirectional recursive structure, extracts the spatial features of smoke diffusion through a self-attention mechanism, and outputs a high-level feature vector containing spatiotemporal correlation information.

[0027] (3) The U-Net decoder receives high-level feature vectors, performs feature dimensionality reduction and dimensionality increase through the encoder-decoder structure, and combines skip connections to fuse the shallow features output by the encoder with the deep features output by the decoder to restore the detailed features of smoke diffusion and output the predicted feature map of smoke diffusion.

[0028] (4) The output layer uses the sigmoid activation function to convert the predicted feature map into a smoke concentration distribution matrix, and outputs the smoke diffusion trajectory vector.

[0029] Step (3) of step 200 of the present invention includes the following steps:

[0030] (1) Coverage calculation: Each grid in the target area is judged one by one. If the predicted smoke concentration ρ of grid j is 1, the coverage is calculated as follows: pred (x j ,y j If ,t)≥ρ0, then the grid is considered to be effectively covered, and the area S of the covered region is included. cover Otherwise, it will not be included.

[0031] After traversing all grids, and combining the spatial attenuation correction coefficient ω, the smoke coverage C(t) at time t is calculated: Among them, S target It is the total area of ​​the target region;

[0032] (2) Coverage blind spot identification: The grid scanning method is used to traverse all grids in the target area. If the predicted smoke concentration ρ of grid j is... pred (x j ,y j If ,t)<ρ0, then the grid is a coverage blind zone; record the coordinate set B=(x b1 ,y b1 ),(x b2 ,y b2 ),...,(x bk ,y bk ), where k is the number of coverage blind spots; at the same time, the area of ​​each coverage blind spot and the required smoke density increment are calculated.

[0033] Step 300 of the present invention includes the following steps:

[0034] Step 301, IA * Improvement of the algorithm's heuristic function: for IA * The algorithm's heuristic function is improved by adding a cooperative conflict penalty term, a coverage improvement gain term, and an energy consumption cost term. The formula for the improved heuristic function is as follows:

[0035] h(n) = φ × h dist (n) + β×h cover (n)-γ×h conflict (n)-δ×h energy (n)-ξ×h wind (n);

[0036] Where h(n) is IA * The heuristic function value of the algorithm, h dist (n) is the distance heuristic term, h cover (n) represents the coverage improvement gain term, h conflict (n) is the cooperative conflict penalty term, h energy (n) is the energy consumption cost term, h wind (n) is the penalty term for flying against the wind, and φ, β, γ, δ, ξ are the weight coefficients of the heuristic function;

[0037] Step 302, Cooperative Scheduling Optimization Process: Taking the set of coordinates of coverage blind spots and smoke screen coverage rate as the core objectives, an improved IA (Integrated Automation) system is adopted. * Algorithm for optimizing collaborative smoke generation scheduling of drone swarms

[0038] Step 303, Optimal scheduling strategy output: After the collaborative scheduling optimization is completed, the optimal collaborative smoke generation scheduling strategy of the UAV cluster is output; when the smoke concentration in the coverage blind area reaches ρ0 or the smoke generation is exhausted, the smoke generation is terminated; the scheduling strategy is transmitted to each UAV in real time through the communication module in the form of instructions to guide the UAV to perform the smoke generation task.

[0039] Step 302 of this invention is specifically divided into three stages:

[0040] (1) First stage: Priority allocation of smoke emission from drones;

[0041] By combining the real-time status of the drones with information on coverage blind spots, a smoke emission priority is assigned to each drone;

[0042] (2) Second stage: UAV smoke emission path planning;

[0043] Starting from the initial position of each drone and ending at the corresponding target coverage blind zone coordinates, the improved IA method is used.* The algorithm plans the optimal smoke-generating path for each drone. During path planning, cooperative conflict detection is performed in real time: the distance between the candidate path of the current drone and the candidate paths of all other drones is calculated. If the distance is less than a preset conflict detection threshold, a cooperative conflict is determined to exist. The cooperative conflict penalty term h in the heuristic function is adjusted accordingly. conflict (n) Re-search for the optimal path until all cooperative conflicts are eliminated; the completed optimal path is output as a sequence of path points, i.e., Path. i ={(x i1 ,y i1 ,z i1 ),(x i2 ,y i2 ,z i2 ),…,(x im ,y im ,z im )}, where m is the number of path points, and the distance between any two adjacent path points is the path search step size. The UAV will fly along the path point sequence to complete the smoke generation task;

[0044] (3) Third stage: Adjustment of smoke output from the drone;

[0045] Based on the required smoke density increment Δρ in the coverage blind zone k The ambient wind speed u, combined with the output smoke distribution ratio, is used to adjust the smoke output q of each drone. i The formula for adjusting smoke output is as follows: ,

[0046] Where k is the smoke emission correction coefficient, Δρ k S is the required smoke density increment for the k-th coverage blind zone. bk It is the area of ​​the kth blind spot, and u is the ambient wind speed. The higher the wind speed, the faster the smoke spreads and the greater the amount of smoke required.

[0047] In the first stage described in this invention, the calculation of the drone's smoke emission priority is as follows:

[0048] ,

[0049] Among them, P i The smoke emission priority of the i-th drone is e. i,max q is the maximum battery power of the i-th drone. i,max d is the maximum smoke emission of the i-th drone. i d is the shortest distance from the current position of the i-th drone to the target's blind spot. max It is the maximum distance from all drones to their respective target coverage blind spots, and ω1, ω2, ω3, and ω4 are priority weight coefficients.

[0050] In the second stage described in this invention, the cooperative conflict detection adopts the minimum distance judgment method. The formula for calculating the minimum distance between the candidate paths of the i-th UAV and the j-th UAV is as follows:

[0051] ,

[0052] Where, d i,j Path is the minimum distance between the candidate paths of the i-th drone and the j-th drone. i It is a candidate path for the i-th drone and consists of multiple path points. j It is a candidate path for the j-th UAV and consists of multiple path points, (x p ,y p ,z p Let (x, y) be the three-dimensional coordinates of any path point p on the candidate path of the i-th UAV, and (x, y) be the coordinates of the path point p on the candidate path of the i-th UAV. q ,y q ,z q ) is the three-dimensional coordinate of any path point q on the candidate path of the j-th UAV;

[0053] The conflict determination logic is: if d i,j < , If the preset threshold for detecting collaborative conflict is used, then a collaborative conflict is determined to exist; otherwise, no collaborative conflict exists.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] 1. This invention applies the IReTUNet model to smoke diffusion prediction. It captures the spatiotemporal correlation features of smoke diffusion through a recursive Transformer structure and restores diffusion details by combining the U-Net structure, thus solving the problems of large prediction bias and poor real-time performance of traditional physical models. At the same time, it introduces a spatial attenuation correction coefficient and a grid scanning method to achieve accurate calculation of smoke coverage and accurate identification of coverage blind spots, providing a reliable decision basis for UAV scheduling optimization.

[0056] 2. This invention adopts IA * The algorithm, by optimizing the heuristic function and incorporating cooperative conflict penalty terms, coverage improvement gain terms, and energy consumption cost terms, achieves collaborative optimization of path planning, smoke generation order allocation, and smoke generation adjustment for drone swarms. This is achieved by integrating IReTUNet with IA... * By combining algorithms, the advantages of deep learning and heuristic search algorithms are complemented, solving the problems of insufficient synergy, easy conflict, and getting trapped in local optima in traditional algorithms, and achieving optimal allocation of smoke generation resources. Attached Figure Description

[0057] Figure 1This is a flowchart illustrating an embodiment of the present invention.

[0058] Figure 2 This is a flowchart illustrating step 200 of an embodiment of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of this application, but not all embodiments.

[0060] See Figure 1 This embodiment provides a method for coordinated smoke generation scheduling of UAV swarms aimed at maximizing smoke screen coverage, including the following steps:

[0061] Step 100: Collect real-time operational data and environmental data from the smoke-generating drone and perform preprocessing. This includes the following steps:

[0062] Step 101: Collect real-time drone operation data: Determine the number N of drones participating in collaborative smoke generation, and collect the current position coordinates, flight speed, remaining battery power, current smoke generation amount, and smoke generation rate of the drones in real time through the GPS positioning module, attitude sensor, and smoke generation sensor carried by each drone.

[0063] Collect environmental data: Collect wind speed, wind direction, and ambient temperature through ground-based environmental monitoring equipment to ensure the real-time nature of the data.

[0064] Step 102: Standardize the real-time UAV operation data and environmental data, mapping all data to the [0,1] interval to eliminate the influence of dimensions. The calculation formula for standardization is:

[0065] ,

[0066] Where, x norm x is the standardized data; x is the original data; x min It is the minimum value of the data; x max It is the maximum value of the data.

[0067] Step 103: Convert the standardized data into a spatiotemporal feature matrix and IA adapted to the input of the IReTUNet model. * (Imperative Learning-based A) * The parameter vector input to the Search algorithm is stored in the database of the ground control station for use in subsequent steps.

[0068] Step 200: Improve the ReTUNet model to obtain the improved ReTUNet model, which is defined as the IReTUNet model. Use the IReTUNet model to predict the smoke screen diffusion state, outputting a smoke screen concentration distribution matrix. Based on this matrix, calculate the smoke screen coverage rate and identify coverage blind spots. Specifically, this includes the following steps:

[0069] Step 201: To address the need for smoke screen diffusion prediction, the traditional ReTUNet model is improved to obtain the improved ReTUNet model, namely the IReTUNet model. The core improvements include two aspects:

[0070] 1. In the recursive Transformer encoder, a spatial decay mechanism based on Manhattan distance is introduced to replace the traditional Transformer self-attention mechanism. When calculating the attention weight between two spatial points, a Manhattan distance decay term is added, as shown in the following formula:

[0071] ,

[0072] Where Attention(i,j) is the attention weight between the i-th spatial point and the j-th spatial point, Sim(i,j) is the feature similarity between the i-th spatial point and the j-th spatial point, α is the spatial decay coefficient, and K is the total number of spatial points; Sim(i,k) is the feature similarity between the i-th spatial point and the k-th spatial point, d man (i,k) is the Manhattan distance between the i-th spatial point and the k-th spatial point. For all exp(Sim(i,k)-α×d from k to k man Summing the terms (i,k) by exp(Sim(i,j)-α×d) man (i,j)) is the exponent of the feature similarity between the i-th spatial point and the j-th spatial point, minus the product of α and the Manhattan distance between them. d man (i,j) is the Manhattan distance between the i-th spatial point and the j-th spatial point, and its calculation formula is: d man (i,j)=|x i -x j |+|y i -y j |, where x i x j These are the x-coordinates of the i-th and j-th spatial points, respectively, and y i y j These are the ordinates of the i-th and j-th spatial points, respectively.

[0073] This improvement can effectively reduce the computational complexity of the self-attention mechanism, while enhancing the model's ability to extract spatial features of smoke diffusion, and better match the spatial decay characteristics of smoke diffusion.

[0074] Second, a multi-scale feature fusion module is added to the U-Net decoder. This module fuses high-level feature vectors of different scales output by the recursive Transformer encoder to restore the detailed features of smoke diffusion and improve prediction accuracy.

[0075] Step 202, Model Training: Train the IReTUNet model. During training, the mean squared error (MSE) is used as the loss function, and the Adam optimizer is used. The model parameters are updated through backpropagation until the model converges. The formula for the mean squared error (MSE) loss function is as follows:

[0076] ,

[0077] Among them, L MSE This is the mean squared error loss function value; the smaller the value, the higher the model's prediction accuracy. M is the number of spatial grids corresponding to each sample, and ρ... pred (i,j) is the predicted smoke concentration of the j-th grid in the i-th sample, ρ true (i,j) represents the actual smoke concentration in the j-th grid of the i-th sample. The trained IReTUNet model can output a predicted feature map of smoke diffusion, which can then be converted into a smoke concentration distribution matrix and a coverage mask.

[0078] Step 203, Smoke screen diffusion prediction: x of the standardized data norm Input the trained IReTUNet model, and the IReTUNet model will perform the prediction process, which includes the following steps:

[0079] (1) The input layer of the IReTUNet model converts the standardized data into a spatiotemporal feature matrix, and the number of features is the number of categories of the input data;

[0080] (2) The recursive Transformer encoder processes the spatiotemporal feature matrix, captures the temporal correlation of smoke diffusion through a bidirectional recursive structure, extracts the spatial features of smoke diffusion through a self-attention mechanism, and outputs a high-level feature vector containing spatiotemporal correlation information.

[0081] (3) The U-Net decoder receives high-level feature vectors, performs feature dimensionality reduction and dimensionality increase through the encoder-decoder structure, and combines skip connections to fuse the shallow features output by the encoder with the deep features output by the decoder to restore the detailed features of smoke diffusion and output the predicted feature map of smoke diffusion.

[0082] (4) The output layer uses the sigmoid activation function to convert the predicted feature map into the predicted smoke concentration of all grids in the corresponding target area, i.e., at time t, the coordinates (x j ,y j The predicted smoke concentration ρ at point ) pred (x j ,y j Simultaneously, the smoke diffusion trajectory vector is output, reflecting the diffusion direction and diffusion rate of the smoke at time t.

[0083] Step 204, Smoke Coverage Calculation and Coverage Blind Spot Identification: Based on the smoke concentration distribution matrix output by the IReTUNet model, combined with the smoke concentration threshold ρ0, the current smoke coverage is calculated, and coverage blind spots are identified. This includes the following steps:

[0084] (1) Coverage calculation: Each grid in the target area is judged one by one. If the predicted smoke concentration ρ of grid j is 1, the coverage is calculated as follows: pred (x j ,y j If ,t)≥ρ0, then the grid is considered to be effectively covered, and the area S of the covered region is included. cover Otherwise, it is not included; after traversing all grids, the smoke coverage C(t) at time t is calculated by combining the spatial attenuation correction coefficient ω: Among them, S target It is the total area of ​​the target region.

[0085] (2) Coverage blind spot identification: The grid scanning method is used to traverse all grids in the target area. If the predicted smoke concentration ρ of grid j is... pred (x j ,y j If ,t)<ρ0, then the grid is a coverage blind zone; record the coordinate set B=(x b1 ,y b1 ),(x b2 ,y b2 ),...,(x bk ,y bk ), where k is the number of coverage blind spots; simultaneously, the area of ​​each coverage blind spot and the required smoke density increment are calculated for subsequent IA * Algorithm scheduling optimization provides a goal-oriented approach.

[0086] Step 205: Combine the smoke coverage C(t) at time t and the predicted smoke concentration ρ. pred (x j ,y j The scheduling optimization module, which outputs the coordinate set B of the blind zone to the ground control station in real time, serves as the IA (Intervention Area Control) module. *The core input for the algorithm to perform cooperative scheduling optimization.

[0087] Step 300, for IA * The algorithm is improved by optimizing its heuristic function; the improved IA is then used. * The algorithm performs collaborative scheduling optimization for drone swarms, completing smoke generation priority allocation, path planning, and smoke generation adjustment; finally, it outputs the optimal scheduling strategy to guide drones in executing smoke generation tasks. Specifically, it includes the following steps:

[0088] Step 301, IA * Improved heuristic function of the algorithm: Traditional IA * The algorithm's heuristic function only considers the path cost of a single UAV, and the improved IA * The algorithm's heuristic function h(n) is modified by adding a cooperative conflict penalty term, a coverage improvement gain term, and an energy consumption cost term to achieve multi-objective cooperative optimization. The improved heuristic function formula is as follows:

[0089] h(n) = φ × h dist (n) + β×h cover (n)-γ×h conflict (n)-δ×h energy (n)-ξ×h wind (n); where h(n) is IA * The heuristic function value of the algorithm; h dist (n) is the distance heuristic; h cover (n) represents the coverage improvement gain term; h conflict (n) is the cooperative conflict penalty term; h energy (n) is the energy consumption cost term; h wind (n) is the penalty for flying against the wind; if the path segment is against the wind, h wind (n) = 1; if the upwind direction covers the blind spot, h wind (n) = 0; φ, β, γ, δ, ξ are the weight coefficients of the heuristic function.

[0090] Step 302, Cooperative Scheduling Optimization Process: Taking the set of blind zone coordinates B and the smoke screen coverage rate C(t) as the core objectives, IA is adopted. * The algorithm optimizes the coordinated smoke generation scheduling of drone swarms, and the specific process is divided into three stages:

[0091] (1) First stage: Priority allocation of smoke generation from drones.

[0092] Combined with the drone's real-time status (remaining battery power e) i Current smoke emission q i Flight speed v iBased on the coverage blind spot information (blind spot area, required smoke density increment), a smoke generation priority P is assigned to each drone. i (i=1,2,…,N), the allocation principle is: drones that are closer to the coverage blind zone, have sufficient remaining power, and have high smoke generation efficiency are given higher priority; areas with large coverage blind zones and large required smoke density increments are given priority to high-priority drones.

[0093] The priority of drone smoke emission is calculated as follows:

[0094] ,

[0095] Among them, P i The smoke emission priority of the i-th drone is e. i,max q is the maximum battery power of the i-th drone. i,max d is the maximum smoke emission of the i-th drone. i d is the shortest distance from the current position of the i-th drone to the target's blind spot. max It is the maximum distance from all drones to their respective target coverage blind spots, and ω1, ω2, ω3, and ω4 are priority weight coefficients.

[0096] (2) Second stage: UAV smoke path planning.

[0097] Starting from the initial position of each drone Starting from the corresponding target coverage blind zone coordinates (x bk ,y bk (The endpoint is ) and the improved IA is adopted. * The algorithm plans the optimal smoke-generating path for each drone. During path planning, cooperative conflict detection is performed in real time: the distance between the candidate path of the current drone and the candidate paths of all other drones is calculated. If the distance is less than a preset conflict detection threshold, a cooperative conflict is determined to exist. The cooperative conflict penalty term h in the heuristic function is adjusted accordingly. conflict (n) Search for the optimal path again until all cooperative conflicts are eliminated; at the same time, the path planning must meet the flight constraints of the UAV and the boundary constraints of the target area to ensure the feasibility of the path.

[0098] The collaborative conflict detection uses the minimum distance judgment method. The formula for calculating the minimum distance between the candidate paths of the i-th UAV and the j-th UAV is as follows:

[0099] ,

[0100] Where, d i,j Path is the minimum distance between the candidate paths of the i-th drone and the j-th drone. i It is a candidate path for the i-th drone and consists of multiple path points.j It is a candidate path for the j-th UAV and consists of multiple path points, (x p ,y p ,z p Let (x, y) be the three-dimensional coordinates of any path point p on the candidate path of the i-th UAV, and (x, y) be the coordinates of the path point p on the candidate path of the i-th UAV. q ,y q ,z q ) is the three-dimensional coordinate of any path point q on the candidate path of the j-th UAV; This represents the candidate path from the i-th drone. i Choose any path point p from the candidate path of the j-th UAV. j Choose any path point q, calculate the three-dimensional Euclidean distance between all path point pairs (p,q) and take the minimum value.

[0101] The conflict determination logic is: if d i,j < , If the preset threshold for detecting collaborative conflict is used, then a collaborative conflict is determined to exist; otherwise, no collaborative conflict exists.

[0102] The planned optimal path is output as a sequence of path points, i.e., Path. i ={(x i1 ,y i1 ,z i1 ),(x i2 ,y i2 ,z i2 ),…,(x im ,y im ,z im )}, where m is the number of path points, and the distance between any two adjacent path points is the path search step size. The UAV will fly along the path point sequence to complete the smoke generation task.

[0103] (3) Third stage: Adjustment of smoke output of drone.

[0104] Based on the required smoke density increment Δρ in the coverage blind zone k The ambient wind speed u, combined with the output smoke distribution ratio, is used to adjust the smoke output q of each drone. i The formula for adjusting smoke output is as follows: ,

[0105] Where k is the smoke emission correction coefficient, Δρ k S is the required smoke density increment for the k-th coverage blind zone. bk It is the area of ​​the kth blind spot, and u is the ambient wind speed. The higher the wind speed, the faster the smoke spreads and the greater the amount of smoke required.

[0106] Step 303, Optimal Scheduling Strategy Output: After the cooperative scheduling optimization is completed, the optimal cooperative smoke scheduling strategy for the UAV swarm is output. This strategy includes the following information for each UAV: ​​smoke priority P i Smoke candidate path i Smoke volume q i Smoke generation termination conditions. Smoke generation terminates when the smoke concentration in the coverage blind zone reaches ρ0 or the smoke generation amount is exhausted. The scheduling strategy is transmitted in real-time to each UAV via the communication module in the form of instructions, guiding the UAV to execute the smoke generation task. This is achieved by combining the IReTUNet model with IA... * The algorithm enables precise scheduling of collaborative smoke generation by drone swarms, effectively solving problems such as low smoke diffusion prediction accuracy, poor coordination, and weak dynamic adaptability in existing technologies. It ensures that the smoke coverage rate approaches the maximum target and adapts to the smoke generation task requirements in various complex scenarios.

[0107] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for coordinated smoke generation scheduling of UAV swarms aimed at maximizing smoke screen coverage, characterized by: Includes the following steps: Step 100: Collect real-time operational data and environmental data of the smoke-generating drone and perform preprocessing; Step 200: Improve the ReTUNet model by defining the improved model as the IReTUNet model. Use the IReTUNet model to predict the smoke diffusion state, output the smoke concentration distribution matrix, and calculate the smoke coverage rate and identify coverage blind spots based on this matrix. The improvements include two aspects:

1. In the recursive Transformer encoder, a spatial decay mechanism based on Manhattan distance is introduced to replace the traditional Transformer self-attention mechanism. When calculating the attention weight between two spatial points, a Manhattan distance decay term is added. Second, a multi-scale feature fusion module is added to the U-Net decoder to fuse high-level feature vectors of different scales output by the recursive Transformer encoder. In step 200, the formula for calculating the attention weight between two spatial points is as follows: , Where Attention(i,j) is the attention weight between the i-th spatial point and the j-th spatial point, Sim(i,j) is the feature similarity between the i-th spatial point and the j-th spatial point, α is the spatial decay coefficient, and K is the total number of spatial points; d man (i,j) is the Manhattan distance between the i-th spatial point and the j-th spatial point, and its calculation formula is: d man (i,j)=|x i -x j |+|y i -y j |, where x i x j These are the x-coordinates of the i-th and j-th spatial points, respectively, and y i y j These are the ordinates of the i-th and j-th spatial points, respectively; Step 300, for IA * The algorithm is improved by optimizing its heuristic function; the improved IA is then used. * The algorithm performs collaborative scheduling optimization for UAV clusters, completing smoke generation priority allocation, path planning, and smoke generation volume adjustment; finally, it outputs the optimal scheduling strategy to guide UAVs in performing smoke generation tasks. Step 300 includes the following steps: Step 301, IA * Improvement of the algorithm's heuristic function: for IA * The algorithm's heuristic function is improved by adding a cooperative conflict penalty term, a coverage improvement gain term, and an energy consumption cost term. The formula for the improved heuristic function is as follows: h(n)=φ×h dist (n)+β×h cover (n)-γ×h conflict (n)-δ×h energy (n)-ξ×h wind (n); Where h(n) is IA * The heuristic function value of the algorithm, h dist (n) is the distance heuristic, h cover (n) represents the coverage improvement gain term, h conflict (n) is the cooperative conflict penalty term, h energy (n) is the energy consumption cost term, h wind (n) is the penalty term for flying against the wind, and φ, β, γ, δ, ξ are the weight coefficients of the heuristic function.

2. The UAV swarm collaborative smoke generation scheduling method for maximizing smoke screen coverage as described in claim 1, characterized in that: Step 100 includes the following steps: Step 101: Collect real-time drone operation data: Determine the number N of drones participating in collaborative smoke generation, and collect the current position coordinates, flight speed, remaining battery power, current smoke generation amount, and smoke generation rate of each drone in real time through the GPS positioning module, attitude sensor, and smoke generation sensor carried by each drone; Collect environmental data: Collect wind speed, wind direction, and ambient temperature through ground environmental monitoring equipment; Step 102: Standardize the real-time operation data and environmental data of the UAV; Step 103: Convert the standardized data into a spatiotemporal feature matrix and IA adapted to the input of the IReTUNet model. * The parameter vector input to the algorithm is stored in the database of the ground control station.

3. The UAV swarm collaborative smoke generation scheduling method for maximizing smoke screen coverage as described in claim 1, characterized in that: Step 200 further includes the following steps: Step (1) Model training: Train the IReTUNet model. During the training process, mean squared error is used as the loss function and Adam optimizer is used. The model parameters are updated through backpropagation until the model converges. Step (2) Smoke diffusion prediction: Input the standardized data into the trained IReTUNet model, and the IReTUNet model performs the prediction process; Step (3) Smoke Coverage Calculation and Coverage Blind Spot Identification: Based on the IReTUNet model, output the smoke concentration distribution matrix, combine it with the smoke concentration threshold, calculate the current smoke coverage, and identify coverage blind spots; Step (4) outputs the current smoke cover coverage, smoke concentration distribution matrix, and coverage blind zone coordinate set to the ground control station in real time as IA * The core input for the algorithm to perform cooperative scheduling optimization.

4. The UAV swarm collaborative smoke generation scheduling method for maximizing smoke screen coverage as described in claim 3, characterized in that: The prediction process of step (2) of step 200 includes the following steps: (1) The input layer of the IReTUNet model converts the standardized data into a spatiotemporal feature matrix, and the number of features is the number of categories of the input data; (2) The recursive Transformer encoder processes the spatiotemporal feature matrix, captures the temporal correlation of smoke diffusion through a bidirectional recursive structure, extracts the spatial features of smoke diffusion through a self-attention mechanism, and outputs a high-level feature vector containing spatiotemporal correlation information. (3) The U-Net decoder receives high-level feature vectors, performs feature dimensionality reduction and dimensionality increase through the encoder-decoder structure, and combines skip connections to fuse the shallow features output by the encoder with the deep features output by the decoder to restore the detailed features of smoke diffusion and output the predicted feature map of smoke diffusion. (4) The output layer uses the sigmoid activation function to convert the predicted feature map into a smoke concentration distribution matrix, and outputs the smoke diffusion trajectory vector.

5. The UAV swarm collaborative smoke generation scheduling method for maximizing smoke screen coverage as described in claim 3, characterized in that: Step (3) of step 200 includes the following steps: (1) Coverage calculation: Each grid in the target area is judged one by one. If the predicted smoke concentration ρ of grid j is 1, the coverage is calculated as follows: pred (x j ,y j If t) ≥ smoke concentration threshold ρ0, then the grid is considered to be effectively covered and the covered area S is included. cover Otherwise, it will not be included. After traversing all grids, and combining the spatial attenuation correction coefficient ω, the smoke coverage C(t) at time t is calculated: Among them, S target It is the total area of ​​the target region; (2) Coverage blind spot identification: The grid scanning method is used to traverse all grids in the target area. If the predicted smoke concentration ρ of grid j is... pred (x j ,y j If ,t)<ρ0, then the grid is a coverage blind zone; record the coordinate set B=(x b1 ,y b1 ),(x b2 ,y b2 ),...,(x bk ,y bk ), where k is the number of coverage blind spots; at the same time, the area of ​​each coverage blind spot and the required smoke density increment are calculated.

6. The UAV swarm collaborative smoke generation scheduling method for maximizing smoke screen coverage as described in claim 5, characterized in that: Step 300 further includes the following steps: Step 302, Cooperative Scheduling Optimization Process: Taking the set of coordinates of coverage blind spots and smoke screen coverage rate as the core objectives, an improved IA (Integrated Automation) system is adopted. * Algorithm for optimizing the coordinated smoke generation scheduling of drone swarms; Step 303, Optimal scheduling strategy output: After the collaborative scheduling optimization is completed, the optimal collaborative smoke generation scheduling strategy of the UAV cluster is output; when the smoke concentration in the coverage blind area reaches ρ0 or the smoke generation is exhausted, the smoke generation is terminated; the scheduling strategy is transmitted to each UAV in real time through the communication module in the form of instructions to guide the UAV to perform the smoke generation task.

7. The UAV swarm collaborative smoke generation scheduling method for maximizing smoke screen coverage as described in claim 6, characterized in that: The specific process of step 302 is divided into three stages: (1) First stage: Priority allocation of smoke emission from drones; By combining the real-time status of the drones with information on coverage blind spots, a smoke emission priority is assigned to each drone; (2) Second stage: UAV smoke emission path planning; Starting from the initial position of each drone and ending at the corresponding target coverage blind zone coordinates, the improved IA method is used. * The algorithm plans the optimal smoke emission path for each drone; During path planning, cooperative conflict detection is performed in real time: the distance between the current UAV's candidate path and the candidate paths of all other UAVs is calculated. If the distance is less than a preset conflict detection threshold, a cooperative conflict is determined to exist, and the cooperative conflict penalty term h in the heuristic function is adjusted accordingly. conflict (n) Re-search for the optimal path until all cooperative conflicts are eliminated; the completed optimal path is output as a sequence of path points, i.e., Path. i ={(x i1 ,y i1 ,z i1 ),(x i2 ,y i2 ,z i2 ),…,(x im ,y im ,z im )}, where m is the number of path points, and the distance between any two adjacent path points is the path search step size. The UAV will fly along the path point sequence to complete the smoke generation task; (3) Third stage: Adjustment of smoke output from the drone; Based on the required smoke density increment Δρ in the coverage blind zone k The ambient wind speed u, combined with the output smoke distribution ratio, is used to adjust the smoke output q of each drone. i The formula for adjusting smoke output is as follows: , Where k is the smoke emission correction coefficient, Δρ k S is the required smoke density increment for the k-th coverage blind zone. bk It is the area of ​​the kth blind spot, and u is the ambient wind speed. The higher the wind speed, the faster the smoke spreads and the greater the amount of smoke required.

8. The UAV swarm collaborative smoke generation scheduling method for maximizing smoke screen coverage according to claim 7, characterized in that: In the first stage, the priority of drone smoke emission is calculated as follows: ， Among them, P i The smoke emission priority of the i-th drone is e. i,max q is the maximum battery power of the i-th drone. i,max d is the maximum smoke emission of the i-th drone. i d is the shortest distance from the current position of the i-th drone to the target's blind spot. max ω1, ω2, ω3, and ω4 are the maximum distances from all drones to their respective target coverage blind spots, and ω4 are priority weighting coefficients; e i v represents the remaining battery power. i This refers to flight speed.

9. The UAV swarm collaborative smoke generation scheduling method for maximizing smoke screen coverage according to claim 7, characterized in that: In the second stage, the cooperative conflict detection uses the minimum distance judgment method. The formula for calculating the minimum distance between the candidate paths of the i-th UAV and the j-th UAV is as follows: ， Where, d i,j Path is the minimum distance between the candidate paths of the i-th drone and the j-th drone. i It is a candidate path for the i-th drone and consists of multiple path points. j It is a candidate path for the j-th UAV and consists of multiple path points, (x p ,y p ,z p Let (x, y) be the three-dimensional coordinates of any path point p on the candidate path of the i-th UAV, and (x, y) be the coordinates of the path point p on the candidate path of the i-th UAV. q ,y q ,z q ) is the three-dimensional coordinate of any path point q on the candidate path of the j-th UAV; The conflict determination logic is: if d i,j < , If the preset threshold for detecting collaborative conflict is used, then a collaborative conflict is determined to exist; otherwise, no collaborative conflict exists.

Images