A UUV swarm mission planning method based on improved ant colony optimization

Abstract

A UUV swarm mission planning method based on improved ant colony optimization belongs to the autonomous control technology for unmanned underwater vehicles (UUVs). It addresses the poor convergence and low solution quality of conventional algorithms for solving mission planning problems for UUV swarms under limited endurance and load constraints. The algorithm initializes various parameters of the ant colony algorithm and distributes the initial pheromone concentration unevenly according to an improved initial pheromone distribution method. A candidate node set is constructed based on the taboo table, the remaining UUV load, and the remaining endurance. The transition probability between nodes is calculated according to an improved state transition rule, and the next node is selected using a roulette wheel. The taboo table and UUV state are updated, and the ant's node path is recorded. The ant determines whether it has traversed all mission points and whether it meets the UUV quantity constraint. The ant's path and path distance are recorded, and the relevant parameters are updated. The iteration is determined to be complete. If so, the pheromone between paths is globally updated according to the improved pheromone update rule. The optimal path is then iterated according to a preset maximum number of iterations. This improvement and optimization of the basic ant colony algorithm accelerates convergence and reduces local optima.

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous control of unmanned underwater vehicles, and relates to a UUV cluster task planning method based on improved ant colony optimization. Background Art

[0002] Unmanned underwater vehicles (UUVs), as small, high-performance underwater intelligent devices, are widely used in military and civilian scenarios, such as military strikes, intelligence reconnaissance, resource exploration, and submarine maintenance. Currently, one of the hot issues in UUV research is swarm mission planning.

[0003] The ant colony algorithm, one of the most commonly used metaheuristic algorithms, has achieved good results in solving cluster task planning problems. However, it still has several drawbacks: 1) The ant colony algorithm's operating mechanism is essentially a positive feedback behavior, which can cause the algorithm to fall into a local optimum to a certain extent and make it difficult to escape autonomously; 2) Its initial pheromone concentration distribution method and pheromone update rules are not targeted. Although they ensure the diversity of the solution set, unnecessary exploration can lead to poor convergence performance of the algorithm;

[0004] 3) The relationship between the randomness and convergence of the algorithm is contradictory. Improving the convergence performance will increase the probability of the algorithm reaching local optimality and reduce the quality of the solution.

[0005] Current research on improving the ant colony algorithm focuses on the following areas: 1) Optimizing the structure of the ant colony algorithm: improving the state transition rules or the logical framework of the optimization algorithm. In "Research on Mobile Robot Path Planning Using Multi-layer Optimized Ant Colony Algorithm," Yang Liwei et al. improved the algorithm's state transition rules, adding distance and smoothness heuristic functions, improving the algorithm's optimization capabilities and convergence speed. In "Optimal Measurement Point Selection for Analog Circuits Based on Dynamic Ant Colony Algorithm," Luo Hui et al. added penalty matrices and distance matrices to the heuristic factors to improve the ant colony's search efficiency. In "Research on Mobile Robot Path Planning Based on Improved Ant Colony Algorithm," Jiang Ming et al. dynamically planned state transitions using adaptive parameters, balancing the algorithm's convergence and randomness. 2) Optimizing the pheromone rule of the ant colony algorithm: optimizing the initial allocation method or pheromone update rule. Li Zhikun et al. optimized the pheromone distribution rules in "Improved variable step size ant colony algorithm for mobile robot path planning", improved the convergence of the algorithm by reducing the influence of local pheromones; Luo Qiang et al. improved the convergence of the algorithm by increasing the pheromone concentration advantage of excellent paths in "Pathplanning of mobile robot based on improved variable step size ant colony algorithm"; Chen Xiong et al. proposed an adaptive coefficient of pheromone volatilization in "An ant colony algorithm for robot path planning", which improved the randomness of the algorithm; 3) Fusion of ant colony algorithm: combining the ideas or processes of the ant colony algorithm with other algorithms. In "Logistics Distribution Path Optimization Based on Improved Ant Colony Algorithm," Zhang Weize et al. combined genetic algorithm operations such as crossover and mutation to improve the algorithm's solution quality. In "Trajectory Planning of an Autonomous Mobile Robot by Evolving Colony System," Wang Lei et al. used a genetic algorithm to provide a basis for initial pheromone distribution, thereby improving the algorithm's convergence. In "Hybrid Ant Colony System Algorithm for Robot Global Path Planning," Lü Jinqiu et al. used the Dijkstra algorithm to plan a suboptimal solution, which was then used as the initial solution for the ant colony algorithm for the next step of planning, thereby improving the solution quality.

[0006] Based on the analysis of the above domestic and foreign literature, the current problems in solving UUV swarm mission planning problems are generally low solution quality, contradictions between randomness and convergence, and weak ability of the algorithm to autonomously escape from local optimality. These problems greatly affect the global optimization ability and convergence performance of the algorithm.

[0007] To this end, the present invention proposes a new UUV cluster mission planning method based on improved ant colony optimization. Summary of the Invention

[0008] In response to the above problems, the technical problem to be solved by the present invention is to provide a high-quality and efficient cluster task planning method, which can reduce the probability of the algorithm falling into local optimality while ensuring the quality of the solution set and the convergence speed, and achieve rapidity and effectiveness in solving the UUV cluster task planning problem.

[0009] To solve the above problems, the present invention provides the following technical solutions.

[0010] A UUV swarm mission planning method based on improved ant colony optimization includes the following steps:

[0011] Step 1: Determine the task point coordinates and load requirements, and obtain the task point distance set based on the task point coordinates;

[0012] Step 2: Initialize the parameters of the ant colony algorithm and implement an unequal distribution scheme for the initial pheromone concentration on the path according to the improved initial pheromone distribution method;

[0013] Step 3: Construct a candidate mission point set based on the taboo table, the remaining load and remaining endurance of the UUV, calculate the state transition probability according to the improved state transition rule, and use roulette to select the next node;

[0014] Step 4: Update the taboo table and UUV status information, and record the ant's node path;

[0015] Step 5: Determine whether the current ant has traversed all task points. If so, determine whether the ant meets the quantity constraint of the UUV, record the ant path and path distance that meets the quantity constraint, and update the relevant influencing factors in the state transition rule;

[0016] Step 6: Determine whether all the ants in this iteration have finished running. If so, the iteration is completed and the pheromone concentration on the path is globally updated according to the improved pheromone update rule. Otherwise, continue to step 3.

[0017] Step 7: Iterate the algorithm according to the preset maximum number of iterations to finally obtain the optimal path for this algorithm operation. The present invention also includes:

[0018] 1) The improved initial pheromone distribution method described in step 2 is used to distribute the initial pheromone concentration of the path unequally, including the following steps:

[0019] At the beginning of the algorithm, the average distance between the current task point and all other task points is calculated. The distances from other task points to this point are compared with the average. Pheromone on paths from task points with distances less than the average are added together, while pheromones on paths with distances less than the average are evaporated additionally. By increasing the difference in pheromone concentration between paths between close and distant nodes, the algorithm's convergence speed can be increased.

[0020] The pheromone initialization formula is as follows:

[0021] τ ij (0) = T0 + Δτ ij (0)

[0022]

[0023] Where τ ij (0) represents the initial pheromone concentration on the path between task points i and j; T0 represents the given initial pheromone concentration (T0 = 5); Δτ ij (0) represents the initial pheromone concentration added to the path between task points i and j. When the distance between task points is less than the average distance, Δτ ij (0) is positive, otherwise it is negative; d ij represents the distance between task points i and j; L represents the number of task points.

[0024] 2) Step 3 and the calculation of the state transition probability of the candidate solution according to the improved state transition rule include the following steps:

[0025] In the improved state transition rule, a new influencing factor "gravity coefficient γ" is innovatively added to the heuristic factor to actively increase the probability of suboptimal nodes being selected in the node selection process according to certain rules, γ∈[0,1).

[0026] A gravitational force γ is added between any two nodes. This gravitational force can reduce the distance between nodes in the state transition probability calculation. As the gravitational force between nodes increases, the distance between the two points decreases, the heuristic factor increases, and the probability of these two nodes being consecutively selected by ants increases. Because local optimality in the early and middle stages of the algorithm has a greater impact on the algorithm's convergence, it is necessary to increase the randomness in the early and middle stages. However, increasing randomness in the later stages of the algorithm will impair the algorithm's convergence performance. In order to simultaneously ensure rapid and stable convergence in the later stages of the algorithm, the gravitational coefficient and its impact on the transition probability are gradually reduced in an adaptive manner as the number of iterations increases. The formula is as follows:

[0027]

[0028]

[0029] Where η ijRepresents the heuristic factor in the basic ant colony algorithm transition probability formula; d ij represents the distance between task points i and j; γ ij represents the gravitational force between task points i and j; χ represents the adaptive coefficient; N represents the current generation; N max Indicates the maximum number of iterations.

[0030] The magnitude of the inter-node attraction, γ, depends on the order in which nodes are selected in the ant's path. Initially, the inter-node attraction is zero. The attraction between two nodes increases as the number of ants that do not select these two nodes increases, following a Poisson distribution. When these two nodes are selected consecutively, the attraction between them returns to zero and begins to change again. The formula is as follows:

[0031]

[0032]

[0033] Where, parameter σ represents the strength of the Poisson distribution, which determines the peak value of the attraction coefficient γ between two points; parameter λ represents the frequency of the Poisson distribution, which determines the frequency at which the attraction coefficient γ between two points reaches its peak value, that is, the number of ants that do not choose these two points consecutively required to reach the peak value; k ij Represents the cardinality of the Poisson distribution, which represents the number of ants that do not choose task points i and j consecutively.

[0034] 3) The pheromone concentration on the path is globally updated according to the improved pheromone update rule in step 6, including the following steps:

[0035] In the improved pheromone update rule, the present invention cancels the pheromone superposition of all ants after each iteration, sorts the ants according to the size of the evaluation value, only superimposes different amounts of pheromones on a certain proportion of ant paths, and punishes a certain proportion of ant paths.

[0036] Innovatively propose the optimal threshold value val for evaluation value best and the worst evaluation value threshold val worst The ants in each generation are classified and given different types of pheromone update amplitudes. The improved pheromone update rules are as follows:

[0037] τ ij (t+1)=(1-ρ)τ ij (t)+Δτ(add) ij

[0038]

[0039] Where, τ ij (t+1) represents the updated pheromone concentration; τij (t) represents the current pheromone concentration; ρ represents the volatility coefficient of pheromone; Δτ(add) ij The concentration of the pheromone that indicates reward or punishment; Δτ ij represents the pheromone concentration accumulated between task points i and j in this generation of ants; σ1~σ5 represent the reward and punishment range of pheromones, where a value greater than 0 indicates additional superposition of pheromones and a value less than 0 indicates additional volatilization of pheromones, where: σ1=8, σ2=2, σ3=1, σ4=-0.25, σ5=-0.75; ant best Ant is the set of excellent ants in this generation, accounting for the top 1 / 4 of the total number of effective ants in this generation; worst is the set of poor ants in this generation, accounting for the last 1 / 4; val ant Represents the evaluation value of the antth ant; val best (i) represents the best evaluation value of the ant of generation i; val best Indicates the optimal threshold of the evaluation value; val worst Indicates the worst threshold value of the evaluation value;

[0040] In the excellent ant set, the evaluation value is better than the optimal threshold val best The ant path is superimposed with more pheromone, and the evaluation value is worse than the optimal threshold val best The ant path only has a small amount of pheromone superposition; in the poor ant set, the evaluation value is worse than the worst threshold val worst The ant path with more additional pheromone volatilization has a better evaluation value than the worst threshold val worst A small amount of additional pheromone volatilization is carried out along the ant paths.

[0041] The optimal evaluation value threshold val best and the worst evaluation value threshold val worst , the assignment rules are as follows:

[0042]

[0043]

[0044] Where iter represents the current number of iterations; val best (i) represents the best evaluation value of the i-th generation ant. When the best evaluation value of the new iteration is better than the optimal threshold of the algorithm, val best Update; val worst (i) is the worst evaluation value of the i-th iteration. It is worth noting that val is only evaluated when the evaluation value of the worst ant in the new generation of ants is still better than the worst evaluation value threshold. worst to update.

[0045] Beneficial effects of the present invention:

[0046] The key technical points of the present invention are as follows: initializing the various parameters of the ant colony algorithm and unequally distributing the initial pheromone concentration of the ant colony algorithm according to the improved initial pheromone distribution method; constructing a candidate node set based on the taboo table, the remaining load of the UUV, and the remaining endurance, calculating the transition probability between nodes according to the improved state transition rule, and selecting the next node using a roulette wheel; updating the taboo table and UUV state, recording the node path of the ant; determining whether the ant has traversed all task points and whether the UUV quantity constraint is met, recording the ant's path and path distance, and updating related parameters; determining whether the current iteration is complete. If so, globally updating the pheromones between paths according to the improved pheromone update rule; iterating according to a preset maximum number of iterations to solve for the optimal path. The present invention improves and optimizes the basic ant colony algorithm to accelerate the algorithm's convergence speed and reduce local optimality.

[0047] The present invention proposes a UUV cluster task planning method based on improved ant colony optimization, which includes the following beneficial effects: by designing an unequal distribution method of the initial pheromone concentration, the invalid search in the early stage of the algorithm is reduced, and the quality and convergence performance of the early solution of the algorithm are improved; the best and worst evaluation value thresholds of the optimization model are innovatively proposed, and by designing its assignment rules, the ants in each generation of ant colonies are classified, and different pheromone concentration update methods are applied, thereby improving the convergence performance of the algorithm while ensuring that the quality of the algorithm's solution is iterated in a better direction; a new influencing factor is innovatively added to the state transition rule, and the ability of the algorithm to autonomously jump out of the local optimum is enhanced by increasing the probability of ants selecting suboptimal nodes when selecting nodes, thereby improving the randomness of the algorithm and the quality of the solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of the UUV cluster mission planning method based on improved ant colony optimization in the present invention;

[0049] Figure 2(a) is a comparison of the effects of the improved initial pheromone distribution rule under the background of the basic ant colony algorithm;

[0050] Figure 2(b) is a comparison of the effects of the improved initial pheromone distribution rule under the background of the improved ant colony algorithm;

[0051] Figure 3(a) is a comparison of the effects of the improved pheromone update rule under the background of the elite ant colony algorithm;

[0052] Figure 3(b) is a comparison of the effects of the improved pheromone update rule under the background of the improved ant colony algorithm;

[0053] Figure 4(a) is a comparison of the effects of the improved state transition rules under the background of the basic ant colony algorithm;

[0054] Figure 4(b) is a comparison diagram of the effects of the improved state transition rules under the background of the improved ant colony algorithm;

[0055] Figure 5 The changing law of the gravitational coefficient in the improved state transfer rule;

[0056] Figure 6 The UUV cluster mission planning scheme provided for the embodiment;

[0057] Figure 7 A comparison chart of the operating results of the basic ant colony algorithm, the elite ant colony algorithm and the algorithm of the present invention provided in the embodiment;

[0058] Figure 8 A comparison chart of the convergence speed of the basic ant colony algorithm, the elite ant colony algorithm and the algorithm of the present invention provided in the embodiment. DETAILED DESCRIPTION

[0059] The specific implementation of the present invention will be further described below with reference to the accompanying drawings and examples.

[0060] The UUV cluster mission planning problem can be described as follows: It is known that a base has K UUVs, and they are heading to known locations in the operating area and the workload demand q i There are a total of L mission points to operate. Each UUV can operate on multiple mission points and needs to return to the base after completing the operation mission.

[0061] A single UUV is limited by workload and endurance (travel distance). The constraints for cluster missions are as follows:

[0062] 1) All mission points require operations;

[0063] 2) Each task point can only be operated once;

[0064] 3) Each UUV performing an operational mission can leave the base at most once;

[0065] 4) The workload of each UUV meets the total demand of all mission points where it performs operations;

[0066] 5) The cruising range of the UUV involved in the operation meets the length of the path it travels between different mission points.

[0067] The optimization goal of the UUV swarm mission planning problem is to minimize the total range of all UUVs. The formula is as follows:

[0068]

[0069] Where Z represents the total range of all UUVs; k represents the number of UUVs participating in the mission; ncar represents the total number of UUVs participating in the survey mission; d ij Represents the distance from task point i to task point j; represents the i-th mission point completed by the k-th UUV participating in the survey mission; n k represents the total number of mission points completed by the kth UUV participating in the survey mission (n k ≠0).

[0070] Example

[0071] For the specific implementation process, see Figure 1 The specific implementation of the UUV cluster mission planning method based on improved ant colony optimization proposed in the present invention includes the following steps:

[0072] Step 1: Determine the task point coordinates and load requirements, and obtain the task point distance set based on the task point coordinates.

[0073] Based on the known coordinates of the starting point and task point, the distance set between the task points can be obtained.

[0074] Step 2: Initialize the parameters of the ant colony algorithm and implement an unequal distribution scheme for the initial pheromone concentration on the path;

[0075] Step 2.1: The initial parameters of the ant colony algorithm are set as follows: 1) Basic parameters of the ant colony algorithm: maximum number of iterations is 100 generations; population size is 50 ants; total pheromone amount per ant Q = 100, initial pheromone concentration T0 = 5; pheromone importance α = 1; heuristic importance β = 3; pheromone volatility coefficient ρ = 0.1; 2) Parameters in the improvement item: Poisson distribution magnification δ = 6; Poisson distribution intensity coefficient λ = 20;

[0076] Step 2.2: Calculate the mean distance between the current task point and all other task points. Compare the distances from other task points to this point with the mean. If the distances are less than the mean, the pheromones along the path from the task point to the current point will be added together; if the distances are less than the mean, additional pheromones will be emitted. By unequally distributing the initial pheromone concentration, the concentrations between closely spaced nodes and those far away will differ. This guides ants to more likely choose nodes closer to the current node in the early stages of the algorithm, reducing the number of iterations and laying a good foundation for rapid convergence.

[0077] The pheromone initialization formula is as follows:

[0078] τ ij (0) = T0 + Δτ ij (0)

[0079]

[0080] Where τ ij (0) represents the initial pheromone concentration on the path between task points i and j; T0 represents the given initial pheromone concentration (T0 = 5); Δτ ij (0) represents the initial pheromone concentration added to the path between task points i and j. When the distance between task points is less than the average distance, Δτ ij (0) is positive, otherwise it is negative; d ij represents the distance between task points i and j; L represents the number of task points.

[0081] Step 3: Construct a candidate node set based on the taboo table, the remaining load and remaining endurance of the UUV, calculate the state transition probability according to the improved state transition rule, and use roulette to select the next node;

[0082] Step 3.1: J k is the set of next candidate task points that meet the constraints. k The candidate task point j in must meet three conditions:

[0083] 1) j∈tabu, where tabu is a taboo table, which is the set of task points that have not yet completed the task;

[0084] 2)Q(t)-q j ≥0, Q(t) represents the remaining working load of UUV at the current moment, q i represents the demand for task point i;

[0085] 3)D(t)-d ij -d j,0 ≥0, D(t) is the remaining endurance distance of the UUV at the current moment.

[0086] when When the UUV will choose to return to the base; when If the UUV's current mission point is not the base, it will also choose to return to the base with a certain probability. If the ant chooses the base as the next node, it will update the workload and range at the base, and then set off again on behalf of the new UUV, and record the number of UUVs represented.

[0087] Step 3.2: In the improved state transition rule, the present invention innovatively adds a kind of gravity γ between any two nodes. The gravity can reduce the value of the distance between nodes in the state transition probability calculation. As the gravity between nodes increases, the distance between the two points becomes smaller, the heuristic factor value becomes larger, and the probability of these two nodes being continuously selected by ants will increase. Because the local optimal situation in the early and middle stages of the algorithm has a greater impact on the convergence effect of the algorithm, it is necessary to improve the randomness in the early and middle stages, and improving the randomness in the late stage of the algorithm will damage the convergence performance of the algorithm. In order to ensure the rapid and stable convergence of the algorithm in the late stage, the gravity coefficient and the impact on the transition probability are gradually reduced in an adaptive manner according to the increase in the number of iterations. The formula is as follows:

[0088]

[0089]

[0090] Where η ij Represents the heuristic factor in the basic ant colony algorithm transition probability formula; d ij represents the distance between task points i and j; γ ij represents the gravitational force between task points i and j; χ represents the adaptive coefficient; N represents the current generation; N max Indicates the maximum number of iterations.

[0091] The magnitude of the inter-node attraction, γ, depends on the order in which nodes are selected in the ant's path. Initially, the inter-node attraction is zero. The attraction between two nodes increases as the number of ants that do not select these two nodes increases, following a Poisson distribution. When these two nodes are selected consecutively, the attraction between them returns to zero and begins to change again. The formula is as follows:

[0092]

[0093]

[0094] Where, parameter σ represents the strength of the Poisson distribution, which determines the peak value of the attraction coefficient γ between two points; parameter λ represents the frequency of the Poisson distribution, which determines the frequency at which the attraction coefficient γ between two points reaches its peak value, that is, the number of ants that do not choose these two points consecutively required to reach the peak value; k ij The cardinality of the Poisson distribution is the number of ants that do not consecutively select task points i and j. Here, δ = 6 and λ = 20.

[0095] Step 4: Update the taboo table and UUV status information, and record the ant's node path;

[0096] When the ant's next node is to return to the base, the UUV's load and cruising distance must be reset, and the number of UUVs already used must be recorded; when the ant's next node is a mission point, the UUV's load and cruising distance must be updated, the ant's node path must be recorded, and the taboo table of the mission point must be updated.

[0097] Step 5: Determine whether the current ant has traversed all task points. If so, determine whether the ant meets the quantity constraint of the UUV, record the ant path and path distance that meets the quantity constraint, and update the relevant influencing factors in the state transition rule;

[0098] Ants are selected based on the number of UUVs used to traverse all task points. The paths with small ants that meet the UUV number limit and the total distance of the paths are recorded. The gravity coefficient γ in the state transition rule is updated based on the order of nodes in the ant path.

[0099] Step 6: Determine whether all the ants in this iteration have finished running. If so, the iteration is completed and the pheromone concentration on the path is globally updated according to the improved pheromone update rule. Otherwise, continue to step 3.

[0100] Step 6.1: In the improved pheromone update rule, the present invention cancels the pheromone superposition of all ants after each iteration, sorts the ants according to the size of the evaluation value, only superimposes different amounts of pheromones on a certain proportion of ant paths, and punishes a certain proportion of ant paths.

[0101] Innovatively propose the optimal threshold value val for evaluation value best and the worst evaluation value threshold val worst The ants in each generation are classified and given different types of pheromone update amplitudes. The improved pheromone update rules are as follows:

[0102] τ ij (t+1)=(1-ρ)τ ij (t)+Δτ(add) ij

[0103]

[0104] Where, τ ij (t+1) represents the updated pheromone concentration; τ ij (t) represents the current pheromone concentration; ρ represents the volatility coefficient of pheromone; Δτ(add) ij The concentration of the pheromone that indicates reward or punishment; Δτ ijrepresents the pheromone concentration accumulated between task points i and j in this generation of ants; σ1~σ5 represent the reward and punishment range of pheromones, where a value greater than 0 indicates additional superposition of pheromones and a value less than 0 indicates additional volatilization of pheromones, where: σ1=8, σ2=2, σ3=1, σ4=-0.25, σ5=-0.75; ant best Ant is the set of excellent ants in this generation, accounting for the top 1 / 4 of the total number of effective ants in this generation; worst is the set of poor ants in this generation, accounting for the last 1 / 4; val ant Represents the evaluation value of the antth ant; val best (i) represents the best evaluation value of the ant of generation i; val best Indicates the optimal threshold of the evaluation value; val worst Indicates the worst threshold value of the evaluation value.

[0105] In the excellent ant set, the evaluation value is better than the optimal threshold val best The ant path is superimposed with more pheromone, and the evaluation value is worse than the optimal threshold val best The ant path only has a small amount of pheromone superposition; in the poor ant set, the evaluation value is worse than the worst threshold val worst The ant path with more additional pheromone volatilization has a better evaluation value than the worst threshold val worst A small amount of additional pheromone volatilization is carried out along the ant paths.

[0106] Step 6.2: Optimal evaluation value threshold val best and the worst evaluation value threshold val worst , the assignment rules are as follows:

[0107]

[0108]

[0109] Where iter represents the current number of iterations; val best (i) represents the best evaluation value of the i-th generation ant. When the best evaluation value of the new iteration is better than the optimal threshold of the algorithm, val best Update; val worst (i) is the worst evaluation value of the i-th iteration. It is worth noting that val is only evaluated when the evaluation value of the worst ant in the new generation of ants is still better than the worst evaluation value threshold. worst to update.

[0110] Step 7: Iterate according to the preset maximum number of iterations to obtain the optimal planning solution for this algorithm run.

[0111] In this embodiment:

[0112] Reference Figure 1 The improved ant colony algorithm in this application is based on the basic ant colony algorithm and has made several improvements, mainly including the initial pheromone distribution method, state transition rules and pheromone update rules. The unequal distribution method of the initial pheromone concentration designed according to the distance between nodes can increase the initial pheromone concentration between nodes with close distances, reduce invalid searches in the early stage of the algorithm, and improve the quality and convergence performance of the algorithm's early solutions; a new influencing factor is added to the state transition rule, which increases the probability of ants selecting suboptimal nodes when selecting nodes, enhances the algorithm's ability to autonomously jump out of the local optimum, and improves the randomness of the algorithm and the quality of the solution; in the pheromone update rule, the best and worst evaluation value thresholds of the optimization model are proposed. By designing its assignment rules, the ants in each generation of the ant colony are classified and different pheromone concentration update methods are applied, thereby improving the convergence performance of the algorithm while ensuring that the quality of the algorithm's solution is iterated in a better direction.

[0113] Referring to Figure 2(a), a control experiment was conducted between a basic ant colony algorithm and a basic ant colony algorithm that incorporated the improved initial pheromone distribution method of the present invention. Both algorithms found the optimal solution, but ultimately failed to converge. This is because the convergence of the ant colony algorithm is essentially the convergence of the pheromone concentration differences between paths, and the pheromone update rules of the basic ant colony algorithm cannot significantly increase the pheromone differences between paths, making it difficult for the algorithm to converge. However, the basic ant colony algorithm with the improved initial pheromone distribution method had significantly lower iteration values in the early stages of the algorithm, and the average iteration value in the later stages of the algorithm was also lower than that of the basic ant colony algorithm. This shows that the optimization of the initial pheromone concentration distribution method in the present invention can improve the quality of the solution in the early stages of the algorithm and also play a positive role in the quality of the algorithm's final solution.

[0114] Referring to Figure 2(b), a comparative experiment was conducted between the improved ant colony algorithm of the present invention and the improved ant colony algorithm without the initial pheromone allocation optimization term. The improved algorithm without the optimization term experienced a higher iteration count in the early stages of the algorithm, resulting in poorer solution quality. This increased the number of generations required for convergence and resulted in a slower convergence rate than the complete improved ant colony algorithm. This indicates that the initial pheromone allocation optimization term accelerates convergence by reducing the iteration count in the early stages of the algorithm.

[0115] Referring to Figure 3(a), a comparative experiment was conducted between the Elite Ant Colony Algorithm (EAA) and an EAC algorithm incorporating the improved pheromone update rule of the present invention. While the EAC algorithm found the optimal solution, it fell into a local optimum due to the lack of any restrictions on the additional pheromone additions to outstanding ants. The EAC algorithm, incorporating the pheromone update optimization, achieved stable convergence after finding the optimal solution. Building on the EAC pheromone update rule, the present invention adds a strategy of "classifying outstanding ants in the current iteration based on their previous best iteration values and then awarding them varying degrees of rewards." This avoids local optimums caused by the additional pheromone additions, improving the quality of the algorithm's solutions.

[0116] Referring to Figure 3(b), a comparative experiment was conducted between the improved ant colony algorithm of the present invention and an improved ant colony algorithm without the pheromone update rule optimization. The improved algorithm without the pheromone update optimization term was able to converge to the optimal value, but the convergence speed was significantly slower than the complete improved algorithm. This is because the pheromone update optimization term eliminates the pheromone superposition for all ants and adds a penalty strategy for low-performing ants, which improves the algorithm's convergence speed.

[0117] Referring to Figure 4(a), a control experiment was conducted between a basic ant colony algorithm and a basic ant colony algorithm with the improved state transition rules of the present invention. The basic ant colony algorithm with the state transition optimization term also failed to converge, but the amplitude of the algorithm's oscillations in the later stages was significantly reduced. This is because the state transition optimization term can increase the probability of suboptimal nodes being selected during the node selection process. This demonstrates that the improvements to the state transition rules of the present invention can limit the increase in algorithm randomness to a reasonable range.

[0118] Referring to Figure 4(b), a comparative experiment was conducted between the improved ant colony algorithm of the present invention and the improved ant colony algorithm with the state transition rule improvement removed. The improved algorithm with the state transition optimization removed fell into a local optimum; the complete improved ant colony algorithm finally found the optimal solution and achieved convergence after escaping two iterative value plateaus. It should be noted that in the second plateau of the complete improved algorithm iteration, the algorithm found two lower generation optimal values in the 35th and 46th generations, and finally actively jumped out of the 15-generation (32nd-47th) plateau, avoiding falling into a local optimum. This is due to the following two reasons: 1. In the optimization of the pheromone update rule of the present invention, the pheromone superposition of the elite ants refers to the optimal value (optimal threshold) of the previous iteration. In the 35th generation, the algorithm found a solution lower than the plateau iteration value. Therefore, in the 36th to 45th generations, the pheromone superposition of the elite ants was carried out according to the minimum standard in the strategy, and the pheromone concentration on the path corresponding to the plateau value was not excessively increased; 2. In the optimization of the state transition rule of the present invention, the algorithm will select suboptimal nodes with a certain probability in node selection, and increase randomness within the good node range to find new excellent solutions (iteration values of the 35th and 46th generations).

[0119] It can be seen that the improvement of the state transition rules is conducive to the algorithm to escape from the local optimum and increase the probability of the algorithm finding the optimal solution.

[0120] refer to Figure 5 ,In the improved state transition rule, the relevant parameters δ=6,λ=20, and the change rule of the gravity coefficient ,is consistent with the characteristics of Poisson distribution, with a peak value of 0.538 and a maximum change period of 40.

[0121] Experimental verification of the present invention:

[0122] Using the example from the paper "Logistics Distribution Path Optimization Based on an Improved Ant Colony Algorithm," a comparative analysis is conducted between the basic ant colony algorithm, the elite ant colony algorithm, the algorithm in the paper, and the improved ant colony algorithm proposed in this paper. When the optimization objective is to minimize the total distance, the consensus regarding the quality of the solution to the task planning problem is that the paths in the plan have no intersection or overlap, which is considered a near-optimal solution.

[0123] refer to Figure 6 When the ant colony algorithm of the present invention is used to solve the cluster task planning problem, the optimal solution obtained does not have path intersection, which meets the definition of an approximate optimal solution in the cluster task planning problem.

[0124] As shown in Table 1, the number of UUVs participating in the operation mission, the endurance and load used by each UUV in the optimal planning scheme meet the constraints of using a maximum of 5 UUVs, a maximum endurance of 50 km for a single UUV, and a maximum load of 8 in the literature.

[0125] Table 1: Planning schemes satisfying capacity constraints

[0126]

[0127] refer to Figure 7 As shown in Table 2, a statistical comparison experiment was conducted on 50 examples from the literature using the basic ant colony algorithm, the elite ant colony algorithm, and the improved ant colony algorithm proposed in this invention. Compared with the basic ant colony algorithm and the elite ant colony algorithm, the improved ant colony algorithm proposed in this invention reduced the average distance of the 50 algorithm runs by 4.138 km and 2.856 km, respectively. The proportion of finding the optimal solution increased by 26% and 24%, respectively, and the proportion of finding the near-optimal solution increased by 78% and 66%, respectively.

[0128] Table 2 Algorithm performance comparison

[0129]

[0130] refer to Figure 8The best results of the three algorithms in 50 runs were selected to compare their convergence performance. The basic ant colony algorithm found the optimal solution of 107.84 km in the 17th generation, but because its pheromone update method could not significantly increase the difference in pheromone concentration between paths, the algorithm did not converge in the end; the elite ant colony algorithm converged in the 54th generation; the improved ant colony algorithm proposed in this invention achieved convergence in the 27th generation.

[0131] The improved ant colony algorithm proposed in this invention was run 10 times, and the statistical results are shown in Table 3, where D t It can be seen from the statistical table that the improved algorithm proposed in the present invention runs 10 times with an average result of 108.6918 km, including 6 optimal solutions of 107.8403 km and 10 near-optimal solutions.

[0132] Table 3 Statistical experimental results of the algorithm of the present invention

[0133]

[0134] Compared with the average distance of 110.3083 km obtained by running the algorithm 10 times in the literature, the average distance, the number of optimal solution statistics and other indicators of the algorithm proposed in the present invention are better than those in the literature.

[0135] The above are only preferred implementation cases of the present invention and are used to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A UUV cluster mission planning method based on improved ant colony optimization, characterized in that: The following steps are involved: Step 1: Determine the task point coordinates and load requirements, and obtain the task point distance set based on the task point coordinates; Step 2: Initialize the parameters of the ant colony algorithm and implement an unequal distribution scheme for the initial pheromone concentration on the path according to the improved initial pheromone distribution method; At the beginning of the algorithm, the average distance between the current task point and all other task points is calculated, and the distances from other task points to this point are compared with the average. If the distance is less than the average, the pheromones on the path between the task points and the current point will be superimposed, otherwise additional pheromones will be emitted; Step 3: Construct a candidate mission point set based on the taboo table, the remaining load and remaining endurance of the UUV, calculate the state transition probability according to the improved state transition rule, and use roulette to select the next node; Innovatively add a new influencing factor "gravity coefficient γ" to the heuristic factor, actively increase the probability of suboptimal nodes being selected in the node selection process according to certain rules, γ∈[0,1); Step 4: Update the taboo table and UUV status information, and record the ant's node path; Step 5: Determine whether the current ant has traversed all task points. If so, determine whether the ant meets the quantity constraint of the UUV, record the ant path and path distance that meets the quantity constraint, and update the relevant influencing factors in the state transition rule; Step 6: Determine whether all the ants in this iteration have finished running. If so, the iteration is completed and the pheromone concentration on the path is globally updated according to the improved pheromone update rule. Otherwise, continue to step 3. Step 7: Iterate the algorithm according to the preset maximum number of iterations to finally obtain the optimal path for this algorithm operation.

2. The UUV cluster mission planning method based on improved ant colony optimization according to claim 1 is characterized in that: Step 2, performing unequal distribution of the initial pheromone concentrations of the paths according to the improved initial pheromone distribution method, includes the following steps: The pheromone initialization formula is as follows: t ij (0)=T0+Δτ ij (0) Where τ ij (0) represents the initial pheromone concentration on the path between task points i and j; T0 represents the given initial pheromone concentration (T0 = 5); Δτ ij (0) represents the initial pheromone concentration added to the path between task points i and j. When the distance between task points is less than the average distance, Δτ ij (0) is positive, otherwise it is negative; d ij represents the distance between task points i and j; L represents the number of task points.

3. The UUV cluster mission planning method based on improved ant colony optimization according to claim 1 is characterized in that: Calculating the state transition probability of the candidate solution according to the improved state transition rule in step 3 includes the following steps: A gravitational force γ is added between any two nodes. The gravitational force can reduce the distance between nodes in the state transition probability calculation. As the gravitational force between nodes increases, the distance between the two points becomes smaller, the heuristic factor value becomes larger, and the probability of these two nodes being continuously selected by ants increases. The local optimal situation in the early and middle stages of the algorithm has a greater impact on the convergence effect of the algorithm, so it is necessary to improve the randomness in the early and middle stages. Increasing the randomness in the late stage of the algorithm will damage the convergence performance of the algorithm. In order to ensure the rapid and stable convergence of the algorithm in the late stage, the gravitational coefficient and its impact on the transition probability are gradually reduced in an adaptive manner according to the increase in the number of iterations. The formula is as follows: Where η ij Represents the heuristic factor in the basic ant colony algorithm transition probability formula; d ij represents the distance between task points i and j; γ ij represents the gravitational force between task points i and j; χ represents the adaptive coefficient; N represents the current generation; N max Indicates the maximum number of iterations; The magnitude of the inter-node gravitational force γ depends on the order in which the nodes are selected in the ant path. Initially, the inter-node gravitational force is 0. The gravitational force between two nodes increases as the number of ants that do not select these two nodes increases, and its law conforms to the Poisson distribution curve. When these two nodes are selected consecutively, the gravitational force between the two points will be reset to zero and start to change again. The formula is as follows: Where, parameter σ represents the strength of the Poisson distribution, which determines the peak value of the attraction coefficient γ between two points; parameter λ represents the frequency of the Poisson distribution, which determines the frequency at which the attraction coefficient γ between two points reaches its peak value, that is, the number of ants that do not choose these two points consecutively required to reach the peak value; k ij Represents the cardinality of the Poisson distribution, which represents the number of ants that do not choose task points i and j consecutively.

4. The UUV cluster mission planning method based on improved ant colony optimization according to claim 1 is characterized in that: The global update of the pheromone concentration on the path according to the improved pheromone update rule described in step 6 includes the following steps: The pheromone superposition of all ants after each iteration is cancelled. Ants are sorted according to the size of the evaluation value. Only a certain proportion of ant paths are superimposed with different amounts of pheromones, and a certain proportion of ant paths are penalized. Innovatively propose the optimal threshold value val for evaluation value best and the worst evaluation value threshold val worst The ants in each generation are classified and given different types of pheromone update amplitudes. The improved pheromone update rules are as follows: t ij (t+1)=(1-ρ)τ ij (t)+Δτ(add) ij Where, τ ij (t+1) represents the updated pheromone concentration; τ ij (t) represents the current pheromone concentration; ρ represents the volatility coefficient of pheromone; Δτ(add) ij The concentration of the pheromone that indicates reward or punishment; Δτ ij represents the pheromone concentration accumulated between task points i and j in this generation of ants; σ1~σ5 represent the reward and punishment range of pheromones, where a value greater than 0 indicates additional superposition of pheromones and a value less than 0 indicates additional volatilization of pheromones, where: σ1=8, σ2=2, σ3=1, σ4=-0.25, σ5=-0.75; ant best Ant is the set of excellent ants in this generation, accounting for the top 1 / 4 of the total number of effective ants in this generation; worst is the set of poor ants in this generation, accounting for the last 1 / 4; val ant Represents the evaluation value of the antth ant; val best (i) represents the best evaluation value of the ant of generation i; val best Indicates the optimal threshold of the evaluation value; val worst Indicates the worst threshold value of the evaluation value; In the excellent ant set, the evaluation value is better than the optimal threshold val best The ant path is superimposed with more pheromone, and the evaluation value is worse than the optimal threshold val best The ant path only has a small amount of pheromone superposition; in the poor ant set, the evaluation value is worse than the worst threshold val worst The ant path with more additional pheromone volatilization has a better evaluation value than the worst threshold val worst The ant paths undergo a small amount of additional pheromone volatilization; The optimal evaluation value threshold val best and the worst evaluation value threshold val worst , the assignment rules are as follows: Where iter represents the current number of iterations; val best (i) represents the best evaluation value of the i-th generation ant. When the best evaluation value of the new iteration is better than the optimal threshold of the algorithm, val best Update; val worst (i) is the worst evaluation value of the i-th iteration. It is worth noting that val is only evaluated when the evaluation value of the worst ant in the new generation of ants is still better than the worst evaluation value threshold. worst to update.

Images