Unmanned aerial vehicle distribution method facing threat target strike

Abstract

The invention belongs to the technical field of unmanned aerial vehicles, and particularly relates to an unmanned aerial vehicle distribution method for threatening target strike. Comprising the following steps: S1, detecting a potential threat target by using a yov7-tiny model, extracting a target feature attribute and performing standardization processing to obtain standardized feature data; s2, constructing a TFWN network, inputting the standardized feature data into the TFWN network for clustering, and determining the number of fuzzy rules of the TFWN network; s3, a BP neural network is used to adjust the connection weight in the TFWN network, the threat degree of the threat target is quantified, and the target strike sequence is determined; and S4, constructing an improved ant colony algorithm, and determining an optimal allocation method for scheduling the unmanned aerial vehicle to execute the corresponding attack task according to the attack sequence of the target. Experiments prove that the improved algorithm is greatly improved in convergence speed and iteration times, and the situation of falling into a local optimal solution is not found in the experiment process.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicles, and particularly relates to a method for allocating unmanned aerial vehicles for striking threat targets. Background Art

[0002] Striking threat targets enables the military to conduct precise and rapid strikes against key nodes and vital parts of the enemy, thus gaining the upper hand in the conflict. However, there are still some problems and challenges in the allocation of unmanned aerial vehicles for striking threat targets. In practical applications, it is limited by the capabilities of data processing and decision-making. Although unmanned aerial vehicles are equipped with advanced sensors and data processing systems, in the face of a large amount of complex data, how to effectively extract, analyze, and make decisions on information remains a challenge.

[0003] A document with the application number "202311447713.8" discloses a method for dynamically allocating tasks for an unmanned aerial vehicle cluster, which is inspired by the prey selection behavior of a coyote group. Considering the dynamic nature of the combat environment and the potential cooperative relationship of the enemy cluster, it preferentially strikes vulnerable targets to improve the combat capabilities and autonomy level of the unmanned aerial vehicle cluster. However, this method does not consider the extraction of target attributes and the analysis and evaluation of target attribute characteristics, lacking the ability to judge targets.

[0004] A document with the application number "202210329745.7" discloses a method for distributed intelligent task allocation of multiple unmanned aerial vehicles for a dynamic environment. This method proposes a method for allocating tasks for an unmanned aerial vehicle cluster based on a dynamic task allocation model, and realizes the real-time expression and decision-making of task allocation by establishing a stochastic game model and a Markov decision model. However, this method also has obvious limitations, that is, it does not fully consider the conditions under the time window constraint. The constraint of the time window is crucial for the selection of task allocation and execution strategies, because it directly affects the scheduling efficiency of unmanned aerial vehicles and the success probability of strike missions. Summary of the Invention

[0005] In view of the above problems, the present invention proposes a method for allocating unmanned aerial vehicles for striking threat targets, aiming to effectively handle the strike missions of threat targets. This method integrates the yolov7-tiny algorithm, the TSK fuzzy wavelet network (Tsk-Fuzzy Wavelet Network, TFWN), the BP neural network, and an improved ant colony algorithm (Ant Colony Optimization, ACO) to achieve the rapid identification, evaluation, and precise strike of threat targets. The technical solution of the present invention is as follows:

[0006] A method for allocating unmanned aerial vehicles for striking threat targets, comprising:

[0007] S1: Use the yolov7-tiny model to detect potential threat targets, extract the target feature attributes and perform normalization processing to obtain normalized feature data;

[0008] S2: Construct a TFWN network, input the normalized feature data into the TFWN network for clustering, and determine the number of fuzzy rules of the TFWN network;

[0009] S3: Use a BP neural network to adjust the connection weights in the TFWN network, quantify the threat level of the threat target, and determine the order of target strikes;

[0010] S4: Construct an improved ant colony algorithm, and determine the optimal allocation method for scheduling drones to perform corresponding strike tasks according to the order of target strikes.

[0011] Preferably, S1 includes:

[0012] S1.1: Send the collected threat target image into the backbone network of the yolov7-tiny model, extract the local features of the image through the convolutional layer and the pooling layer, and output the feature map;

[0013] S1.2: Upsample the feature map output by S1.1 to match the resolution of the high-level feature map, then adjust the depth of the high-level feature map through a 1×1 convolution, splice and fuse the upsampled feature map and the adjusted high-level feature map in the channel dimension, and further convolve the fused feature map to finally output a feature map containing multi-scale semantic and spatial information;

[0014] S1.3: In the head network of the yolov7-tiny model, perform target attribute detection on the feature map extracted by S1.2, and output the type, status, speed, distance, attack angle, and combat environment situation to obtain target feature information;

[0015] S1.4: Normalize the target feature information obtained in S1.3 through the min-max normalization method, define the range interval to [0,1], and obtain the normalized feature data.

[0016] Preferably, S2 includes:

[0017] S2.1: Improve the TSK fuzzy network, and then construct a TFWN network;

[0018] S2.2: Cluster the target feature information collected in S1.3, and the number of clusters is the number of fuzzy rules of the TFWN network.

[0019] Preferably, S4 includes:

[0020] S4.1: Improve the ant colony algorithm, set the ACO pheromone, including setting the initial pheromone τ in ACO ij , set the maximum number of iterations S and the evaporation rate ρ of the pheromone, and introduce a chaotic particle into the ACO;

[0021] S4.2: According to the strike target order determined in S3, allocate drones to the first target entering the strike range;

[0022] S4.3: Update the drone list according to the time window constraint, and let the k-th drone decide the drone to strike the next target according to the selection probability;

[0023] S4.4: After completing the entire drone-target allocation for all strike targets, calculate the fitness of the corresponding path of each individual and compare it with the fitness of other drones;

[0024] S4.5: According to the pheromone increment Δτ ij Determine whether the current allocation method is the optimal solution based on the comparison result. If the calculated current fitness is the best, the current solution is considered the optimal solution; otherwise, it is not the optimal solution. This optimal solution is the best allocation relationship between the drones and the targets. If it is the optimal solution, jump to S4.6; if it is not the optimal solution, update the pheromone according to the following formula:

[0025] τ ij (t + Δt) = (1 - ρ)τ ij (t) + Δτ ij

[0026]

[0027] where τ ij (t) is the pheromone in ACO, 0 < ρ ≤ 1 represents the evaporation rate of the pheromone, and Δτ ij represents the pheromone increment left on the path [i, j], S represents the current iteration number, and S max represents the maximum number of iterations set in S4.1, ρ max and ρ min are the set evaporation rate thresholds, f best is the fitness corresponding to obtaining the optimal solution, and f sub is the fitness corresponding to obtaining the sub-optimal solution;

[0028] Judge whether the updated pheromone can calculate the optimal solution. If it can, perform S4.6; if it cannot calculate the optimal solution, continue to update the pheromone;

[0029] S4.6: Output the calculated optimal solution, which is the best allocation relationship between the drones and the targets.

[0030] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0031] 1) An innovative improvement has been made to the TSK fuzzy network. By introducing a wavelet neural network to replace the conclusion part derived from fuzzy rules, the accuracy of target threat assessment has been effectively improved. The addition of the wavelet neural network enhances the ability to capture target features, making the evaluation results more accurate.

[0032] 2) By improving the transfer probability of individuals and the parameter update method of the traditional ACO and introducing a chaotic particle, an improved ACO is obtained. The improved ACO is used to solve the UAV - target allocation problem proposed by the present invention. Through multiple experiments, it is found that compared with the traditional ACO, the algorithm of the present invention has a significant improvement in the iteration speed and reduces the probability of the algorithm falling into local optimum. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, but do not constitute a limitation to the present invention.

[0034] In the drawings:

[0035] Figure 1 is the threat target allocation and strike flowchart;

[0036] Figure 2 is the TFWN network structure diagram;

[0037] Figure 3 is the threat assessment algorithm flowchart;

[0038] Figure 4 is the threat assessment algorithm training flowchart;

[0039] Figure 5 is the improved ant colony algorithm flowchart. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] The following is a description of the preferred embodiments of the present invention in conjunction with the attached Figure 1 - attached Figure 5 It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0041] Embodiment:

[0042] Such as Figure 1The following is the threat target allocation and strike flow chart of the present invention. First, the yolov7-tiny algorithm is used to accurately detect potential threat targets and extract the characteristic attributes of the targets. Subsequently, TFWN is used to cluster the detected target characteristic attributes and determine the number of fuzzy rules. Next, the BP neural network is used to determine the optimal parameters of TFWN, complete the quantification of the threat level of each target, and determine the order of target strikes. Finally, according to the order of target strikes, the drones are intelligently scheduled to perform corresponding strike tasks.

[0043] A method for allocating drones for threat target strikes, including:

[0044] S1: Use the yolov7-tiny model to detect potential threat targets, extract the target characteristic attributes and perform normalization processing to obtain normalized characteristic data. It includes:

[0045] S1.1: Send the collected threat target image into the backbone network of the yolov7-tiny model, extract the local features of the image through the convolutional layer and pooling layer, and output the feature map;

[0046] S1.2: Upsample the feature map output by S1.1 to match the resolution of the high-level feature map, then adjust the depth of the high-level feature map through 1×1 convolution, splice and fuse the upsampled feature map and the adjusted high-level feature map in the channel dimension, and further convolve the fused feature map to finally output a feature map containing multi-scale semantic and spatial information;

[0047] S1.3: In the head network of the yolov7-tiny model, perform target attribute detection on the feature map extracted by S1.2, and output the type, status, speed, distance, attack angle, and combat environment situation to obtain target feature information;

[0048] S1.4: Normalize the target feature information obtained in S1.3 through the min-max normalization method, define the range interval to [0,1], and obtain the normalized characteristic data.

[0049] S2: Construct a TFWN network, input the S1 normalized characteristic data into the TFWN network for clustering, and determine the number of fuzzy rules of the TFWN network. It includes:

[0050] S2.1: Improve the TSK fuzzy network, and then construct a TFWN network.

[0051] The TSK fuzzy network describes the relationship between inputs and outputs through a set of fuzzy logic rules. It combines the intuitiveness of fuzzy logic and is applicable to various complex modeling and prediction tasks. However, due to the diversity and complexity of target feature information, a wavelet neural network is introduced based on the TSK fuzzy network to replace the conclusion part derived from the fuzzy rules, and the TFWN network is designed, enhancing the multi-scale analysis ability of target information and the capture ability of target features, making the evaluation results more accurate.

[0052] The constructed TFWN network model is shown as follows:

[0053] R j :IF x1 is A 1j AND...x i is A ij THEN u j is v j =ω j ψ j

[0054] In the formula, x1 is the first input of the TFWN network model, and A 1j is the fuzzy set of the j-th rule of x1 with the fuzzy membership function of , x i is the i-th input of the TFWN network model (i = 1, 2,..., n), and A ij is the fuzzy set of the j-th rule of x i with the fuzzy membership function of , u j is the output of the sixth layer of the TFWN network, v j is the input from the fifth layer to the sixth layer of the TFWN network, ψ j represents the output of the fifth layer of the TFWN network, and ω j represents the connection weight between the fifth layer and the sixth layer of the TFWN network.

[0055] As Figure 2 shown, after the data is input into the TFWN network model, it is directly passed backward by the neurons in the first layer, and the membership function is used to calculate the input data in the second layer:

[0056]

[0057] In the above formula, c ij and σ ij both represent the membership function parameters of the j-th rule.

[0058] After the input data is calculated by the membership function in the second layer, it is input into the third layer, where there are a total of R1 - R RThere are a total of R nodes, and the number of nodes represents the number of fuzzy rules. The output of each node is as follows:

[0059]

[0060] The nodes in the fourth layer are composed of a part of the wavelet neural network. The number of nodes in this layer is the product of the number of nodes in the input layer and the number of rules, that is, n×R. The output of each node is calculated by translating and scaling the mother wavelet of the activation function for the input data, as shown in the following formula:

[0061]

[0062] where t ij and d ij represent the translation and scaling coefficients of the mother wavelet function of the activation function respectively. Among them, u ij (k) is calculated as shown in the following formula:

[0063]

[0064] In the formula, θ ij is the parameter of the wavelet transform, and φ ij (k - 1) is the mother wavelet activation function.

[0065] The output of each node in the fourth layer is passed to the fifth layer according to different corresponding rules. Each node in the fifth layer represents a rule, and the output formula of the fifth layer is as shown in the following formula:

[0066]

[0067] The output of each node in the sixth layer is the product of the output of the fifth layer and the output of the nodes in the third layer, which is calculated by the following formula:

[0068]

[0069] where v j (k) is the input of the data corresponding from the fifth layer to the sixth layer:

[0070]

[0071] In the formula, ω j represents the connection weight between the fifth layer and the sixth layer.

[0072] It is obtained by the following formula:

[0073]

[0074] The total output result of the entire TFWN network is the accumulation of the outputs of each node in the sixth layer, which is calculated by the following formula.

[0075]

[0076] S2.2: Cluster the target feature information collected in S1.3, and the number of clusters is the number of fuzzy rules of the TFWN network;

[0077]

[0078] Among them, ρ k represents the number of other data points in the data point set of the threat target feature information whose distance from our reference object is less than the distance threshold d c , and N represents the total number of data points in the data set. If μ l = 1, it is considered that all data is included under this clustering condition; if μ l < 1, it means that there must be data that does not belong to any class; if μ l > 1, it means that the sum of the number of data in each class is greater than the total number of data, that is, some classes contain the same data points;

[0079] After the clustering algorithm, T n classes are obtained, and the number of fuzzy rules of the TFWN network is T n .

[0080] S3: Use the BP neural network to adjust the connection weights in the TFWN network, quantify the threat level of the threat target, and determine the order of target strikes.

[0081] The update formula for the connection weight ω j is as follows:

[0082]

[0083] ω j (k + 1) represents the updated connection weight, ω j (k) represents the connection weight before update, and γ ω (k) represents the initial connection weight.

[0084] Determine the optimal connection weight according to the relationship between the connection weight and the number of iterations and convergence speed of the TFWN network.

[0085] For example Figure 5 . The judgment criterion for the optimal learning rate is that when the TFWN reaches the minimum number of iterations and the fastest convergence speed, the corresponding connection weight is the optimal. Specifically, too low a connection weight will lead to a slow iteration process and low convergence efficiency; too high a connection weight may increase the risk of falling into a local optimal solution, and when the connection weight further increases to an unreasonable range, the network may not even converge. Therefore, accurately selecting the connection weight plays a decisive role in ensuring the effective convergence of the network.

[0086] In the initial stage of network training, the present invention trains the network with the lowest connection weights, and exponentially increases the connection weights successively as the training progresses. A loss graph is drawn based on the output results of the TFWN network to select the optimal connection weights. According to the determined optimal connection weights, the threat level of the target is output using the TFWN network, that is, the output results of the TFWN network after determining the optimal connection weights are: [0], (0, 0.2], (0.2, 0.4], (0.4, 0.6], (0.6, 0.8], (0.8, 1.0]. According to the output results, the threat level of the target is divided into six levels: none, extremely low, low, medium, high, and extremely high, and the order of target strikes is determined according to the level of threat.

[0087] S4: Construct an improved ant colony algorithm (ACO), and determine the optimal allocation method for scheduling drones to perform corresponding strike tasks according to the order of target strikes. It includes:

[0088] S4.1: Improve the ant colony algorithm, set the ACO pheromone, including setting the initial pheromone τ i j in ACO, setting the maximum number of iterations S and the evaporation rate ρ of the pheromone, and introducing a chaotic particle into the ACO to solve the problem that the traditional ACO randomly initializes the initial value, which may cause the algorithm to fall into a local optimum. The mathematical representation of this particle is as follows:

[0089] X i+1 = μX i (1 - X i )

[0090] In the formula, the value range of X i is (0, 1), μ is the chaotic parameter. To ensure that the value range of X i+1 is the same as the value range of X i , when X i = 0.5, X i+1 takes the maximum value and the maximum value is 1, then the value range of μ is (0, 4]. Here, let μ = 4.

[0091] The formula for the selection probability before adding the chaotic particle is as follows:

[0092]

[0093] The formula for the selection probability after adding the chaotic particle is as follows:

[0094]

[0095] Among them, the transition probability represents that at time t, drone k departs from v i and goes to vj The probability, τ ij (t) is the pheromone content of the path [v i , v j , η ij is the visibility, and α and β are the weights of the ant colony algorithm for pheromone content and heuristic information respectively. is the set of destination nodes that the UAV k can reach next.

[0096] S4.2: According to the order of the strike targets determined in S3, allocate UAVs to the first target that enters the strike range;

[0097] That is, when the enemy enters our early warning range, strike allocation should be carried out. Assume that the i-th UAV in our UAV list is at t ij strikes the target with the serial number j in the enemy threat target group, then t ij must satisfy that it is before the enemy target enters the dangerous area of our key protected objects and ensure that our UAV can effectively strike it at this moment. This constraint condition can be expressed mathematically as:

[0098] t fij ≤t ij ≤t fij +t sij

[0099] In the formula, t fij is the moment when the target with the serial number j in the enemy threat target group reaches the maximum range of the i-th UAV in our UAV list; t sij is the time when the target with the serial number j in the enemy threat target group reaches the dangerous area of our key protected target by the UAV.

[0100] S4.3: Update the UAV list according to the time window constraint, and let the k-th UAV determine the UAV for the next target strike according to the selection probability; the selection probability formula is as shown in the formula of the selection probability after adding chaotic particles in S4.1.

[0101] S4.4: After completing the entire UAV-target allocation for all strike targets, calculate the fitness of each individual's corresponding path and compare it with the fitness of other UAVs;

[0102] The fitness calculation formula is as follows:

[0103]

[0104] τ ij (t) is the pheromone concentration in the ACO, and α is the parameter of the importance of pheromone, which determines the influence degree of pheromone on the path selection of ants. η ijis heuristic information, usually inversely proportional to the distance between v i and v j That is, d ij represents the distance between v i and v j β is the importance parameter of heuristic information, which determines the influence degree of distance on the path selection of ants.

[0105] S4.5: According to the pheromone increment Δτ ij judge whether the current allocation method is the optimal solution by comparing the results. If the calculated current fitness is the best, the current solution is considered the optimal solution; otherwise, it is not the optimal solution. This optimal solution is the best allocation relationship between the UAV and the target. If it is the optimal solution, jump to S4.6. If it is not the optimal solution, update the pheromone according to the following formula:

[0106] τ ij (t + Δt) = (1 - ρ)τ ij (t) + Δτ ij

[0107]

[0108] Among them, 0 < ρ ≤ 1 represents the evaporation rate of pheromone, Δτ ij represents the pheromone increment left on the path [i, j], S represents the current iteration number, S max represents the maximum iteration number set in S4.1, ρ max and ρ min are the set evaporation rate thresholds, f best is the fitness corresponding to obtaining the optimal solution, f sub is the fitness corresponding to obtaining the sub-optimal solution.

[0109] Judge whether the updated pheromone can calculate the optimal solution. If it can, perform S4.6. If it cannot calculate the optimal solution, continue to update the pheromone.

[0110] S4.6: Output the calculated optimal solution, which is the best allocation relationship between the UAV and the target.

[0111] For example, number the enemy threat targets and sort them in the order of their entry into our side as {4, 3, 5, 1, 2}. The solution obtained by using the improved ACO algorithm proposed by the present invention, that is, the optimal strike allocation, is {4, 1, 3, 3, 4, 4, 2, 2, 3, 1}. Subsequently, according to this allocation plan, the UAV will perform corresponding strike tasks to achieve the efficient achievement of the combat goal.

[0112] Table 1 shows the performance comparison between the improved ACO and the traditional ACO. It can be seen from Table 1 that both the traditional ACO and the improved ACO can reach the same optimal solution. However, in 50 simulation experiments, the average number of iterations of the improved ACO is only 10 times, which is significantly lower than the 42 times required by the traditional algorithm. This result indicates that the improved algorithm significantly improves the efficiency of finding the optimal solution, which is very beneficial for improving the real-time response ability of the battlefield environment. In addition, the maximum number of iterations of the improved ACO is 13 times, which is not much different from the average number of iterations, while the traditional ACO does not converge during the calculation process. The improvement in the iteration speed indicates that the stability of the algorithm in the actual combat scenario has been significantly improved, thus enhancing the reliability of the system.

[0113] Table 1 Comparison Table of Improved ACO and Traditional ACO

[0114]

[0115] Generally speaking, in the scenario discussed in the present invention, the use of the improved ACO can significantly improve the efficiency and reliability of UAV allocation. This method can quickly and stably provide the optimal UAV strike assistance decision-making scheme in the battlefield environment, and has certain practical application value.

[0116] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for allocating unmanned aerial vehicles for striking threat targets, characterized in that: Including: S1: Use the yolov7-tiny model to detect potential threat targets, extract the target feature attributes and perform normalization processing to obtain normalized feature data; S2: Construct a TFWN network, input the normalized feature data into the TFWN network for clustering, and determine the number of fuzzy rules of the TFWN network; S3: Use a BP neural network to adjust the connection weights in the TFWN network, quantify the threat level of the threat target, and determine the order of target strikes; S4: Construct an improved ant colony algorithm, and determine the optimal allocation method for scheduling drones to perform corresponding strike tasks according to the strike order of the targets.

2. The method for allocating unmanned aerial vehicles for striking threat targets according to claim 1, wherein: S1 includes: S1.1: Send the collected threat target image into the backbone network of the yolov7-tiny model, extract the local features of the image through the convolutional layer and the pooling layer, and output the feature map; S1.2: Upsample the feature map output by S1.1 to match the resolution of the high-level feature map, then adjust the depth of the high-level feature map through a 1×1 convolution, splice and fuse the upsampled feature map and the adjusted high-level feature map in the channel dimension, and further convolve the fused feature map to finally output a feature map containing multi-scale semantic and spatial information; S1.3: In the head network of the yolov7-tiny model, perform target attribute detection on the feature map extracted by S1.2, and output the type, status, speed, distance, attack angle, and combat environment situation, to obtain the target feature information; S1.4: Normalize the target feature information obtained in S1.3 by the min-max normalization method, and define the range interval to [0,1] to obtain the normalized feature data.

3. The method for allocating unmanned aerial vehicles for striking threat targets according to claim 2, characterized in that: S2 includes: S2.1: Improve the TSK fuzzy network, and then construct a TFWN network; S2.2: Cluster the target feature information collected in S1.3, and the number of clusters is the number of fuzzy rules of the TFWN network.

4. The method for allocating unmanned aerial vehicles for striking threat targets according to claim 3, characterized in that: S4 includes: S4.1: Improve the ant colony algorithm, set the ACO pheromone, including setting the initial pheromone τ in ACO ij , set the maximum number of iterations S and the evaporation rate ρ of the pheromone, and introduce a chaotic particle into ACO; S4.2: According to the strike target order determined in S3, allocate drones to the first target entering the strike range; S4.3: Update the drone list according to the time window constraint, and let the kth drone determine the drone for the next target strike according to the selection probability; S4.4: After all strike targets have completed the entire drone-target allocation, calculate the fitness of the corresponding path of each individual and compare it with the fitness of other drones; S4.5: According to the pheromone increment Δτ ij Based on the comparison result, determine whether the current allocation method is the optimal solution. If the calculated current fitness is the best, then consider the current solution as the optimal solution; otherwise, it is not the optimal solution. This optimal solution is the best allocation relationship between the UAV and the target. If it is the optimal solution, jump to S4.6; if not, update the pheromone according to the following formula: τ ij (t + Δt) = (1 - ρ)τ ij (t) + Δτ ij Among them, τ ij (t) is the pheromone in ACO, 0 < ρ ≤ 1 represents the evaporation rate of pheromone, and Δτ ij represents the increment of pheromone left on the path [i, j], S represents the current iteration number, and S max represents the maximum number of iterations set in S4.1, ρ max and ρ min are the set evaporation rate thresholds, f best is the fitness corresponding to obtaining the optimal solution, and f sub is the fitness corresponding to obtaining the sub-optimal solution; Judge whether the updated pheromone can calculate the optimal solution. If it can, perform S4.

6. If the optimal solution cannot be calculated, continue to update the pheromone; S4.6: Output the calculated optimal solution, which is the best allocation relationship between the drones and the targets.

Images