Complex task planning method based on improved wolf pack algorithm

Abstract

The invention discloses a complex task planning method based on an improved wolf pack algorithm. Chaos reverse learning initialization is carried out on the wolf pack after parameter initialization; calculating target function value sorting of the wolf individuals to select a first wolf, and then selecting a probing wolf as a fierce wolf; judging whether the maximum number of iterations is reached, if so, outputting an optimal solution, otherwise, turning to A; the method comprises the following steps: A, sending a migration command by a head wolf, judging whether the detected wolf reaches the maximum migration frequency, if so, turning to B, otherwise, continuing to migrate, and if the target function value of the detected wolf is superior to that of the head wolf during migration, replacing the head wolf, stopping migration and turning to B; b, the head wolf summons a calling command, the fierce wolf responds to the head wolf, if it is judged that the distance between the head wolf and the fierce wolf is smaller than a distance threshold value, the step C is executed, and otherwise, calling continues; if the summoning medium-fierce wolf objective function is superior to the first wolf, replacing the first wolf, and turning to C; c, the first wolf sends a purse instruction, and all wolf individuals except the first wolf respond to purse; and calculating a wolf pack objective function value to eliminate poorer individuals in the wolf pack. According to the method, the task allocation resolving efficiency and correctness are improved.

Description

Technical Field

[0001] The present invention relates to the field of multi-agent system / unmanned aerial vehicle technology, and in particular to a complex task planning method based on an improved wolf pack algorithm. Background Art

[0002] With the development of aviation technology and computer information technology, various advanced new drones have emerged. In the prior art, the task results of drone planning are highly dependent on the initialization results of the population. When the number of drones to be planned is small, the number of tasks is small, and the battlefield constraints are relatively large, this method can give an acceptable solution. However, with the increasing number of tasks on modern battlefields and the transformation of drones from single drones to clusters, the complexity of the battlefield has led to a significant increase in the number of feasible solutions for task planning, while the population of intelligent algorithms is often small. In this case, a small number of populations are randomly mapped to the entire feasible solution space, which is likely to cause population aggregation and insufficient population diversity. All optimization measures are limited to a piece of solution space, and the entire solution space cannot be explored. In this case, the results of drone task planning are highly dependent on the position of population initialization. If it happens to be in the optimal solution area, the algorithm can find a very good local optimal solution, but the population position deviates far from the optimal solution, and the algorithm will lose the opportunity to find the optimal solution forever. In addition, in the prior art, the exploration step size of the population is fixed each time. As a result, when the target point is far away, more iterations are required to reach the target point. When the population is trapped in a local optimal solution, the local solution range is too large due to the small step size, which makes it impossible to jump out of the local optimal solution. Take a two-dimensional function as an example. Figure 2 As shown, the individual is initially located at xxx and moves one square at a time. To move to the optimal solution, X iterations are required. However, when reaching point P (local optimal point), since the moving step length each time is only 1, it can never jump out of the local optimal range and cannot find the global optimal solution. Summary of the invention

[0003] The purpose of the present invention is to provide a complex task planning method based on an improved wolf pack algorithm, which greatly improves the efficiency and accuracy of UAV task allocation solution.

[0004] Technical solution: A complex task planning method based on an improved wolf pack algorithm, comprising the following steps:

[0005] Step 1, initialize the wolf pack algorithm parameters;

[0006] Step 2, perform chaos reverse learning initialization on the wolf pack initialized in Step 1;

[0007] Step 3, calculate the objective function values ​​of the individual wolves in the wolf pack, and sort the objective values ​​to select the alpha wolf, and then select the scout wolves of the wolf pack according to the scout wolf ratio factor, and the remaining wolves become fierce wolves;

[0008] Step 4, determine whether the maximum number of iterations Maxgon is reached. If so, output the optimal solution, otherwise go to Step 5;

[0009] Step 5, the alpha wolf issues a wandering command to determine whether the scout wolf has reached the maximum wandering times. If so, go to Step 6, otherwise continue wandering. If the scout wolf's objective function value is better than the alpha wolf during wandering, the scout wolf replaces the alpha wolf as the alpha wolf of the wolf pack and immediately stops wandering, and go to Step 6;

[0010] Step 6, the alpha wolf issues a call command, and the fierce wolves in the wolf pack respond to the command and rush towards the alpha wolf, and determine whether the distance between the alpha wolf and the fierce wolves is less than the distance threshold. If so, go to Step 7, otherwise continue the call behavior; if the fierce wolf has a better objective function than the alpha wolf in the call behavior, the fierce wolf replaces the alpha wolf as the alpha wolf of the wolf pack and goes to Step 7, otherwise continue the call behavior until the distance with the alpha wolf is less than the distance threshold;

[0011] Step 7: The alpha wolf issues a siege command, and all wolves except the alpha wolf respond to the command and engage in siege behavior;

[0012] Step 8, calculate the objective function value of the wolf pack, eliminate the worse individuals in the wolf pack according to the inferior wolf elimination mechanism, update the wolf pack and then go to Step 2.

[0013] In Step 1 of the aforementioned complex task planning method based on the improved wolf pack algorithm, the parameters of the wolf pack algorithm are initialized, including: wolf pack size N, maximum number of iterations Maxgon, wolf exploration ratio factor, maximum number of wandering times, summoning distance threshold, wolf pack update ratio β, summoning step length, and siege step length.

[0014] In the aforementioned complex task planning method based on the improved wolf pack algorithm, the summoning step length and the siege step length decrease successively.

[0015] In Step 2 of the aforementioned complex task planning method based on the improved wolf pack algorithm, the tent chaotic map is used to evenly distribute the wolf pack in the entire solution space when the chaotic reverse learning is initialized. At the same time, the idea of ​​reverse learning is used to further expand the wolf individuals in the wolf pack and enrich the diversity of the wolf individuals.

[0016] In Step 2 of the aforementioned complex task planning method based on the improved wolf pack algorithm, the process of initializing the chaotic reverse learning is as follows:

[0017] Construct the equation for the Tent sequence:

[0018]

[0019] In the formula, x n is the element value of the nth generation tent sequence, x n+1 For the next generation, rand(0,1) is a random function

[0020] Let the total number of tasks be T num , the total number of drones is U num , all possible individual wolf task allocation matrices from 1 to Number them in sequence; take the generated chaotic sequence value x i , the matrix number No represented by it is calculated as follows:

[0021]

[0022] Through the matrix number No, inversely deduce the matrix represented by No;

[0023] UAV i The task allocation matrix is ​​x i , the serial number of the drone assigned to task t is The total number of drones is U num , then for Its inverse value is:

[0024]

[0025] In Step 3 of the aforementioned complex task planning method based on the improved wolf pack algorithm, the objective function value is the weighted average of various indicators of concern after the drone allocation scheme is selected.

[0026] In Step 5 of the aforementioned complex task planning method based on the improved wolf pack algorithm, wandering is the intelligent behavior of the exploratory wolf, which is an individual exploring in the wolf pack, and its step length is generated by using the variable step length of the Levy's Fight algorithm.

[0027] In Step 5 of the aforementioned complex task planning method based on the improved wolf pack algorithm, the wandering process is as follows:

[0028] The Mentegna algorithm is used to simulate the search process of Levy's Fight. The step length calculation formula is:

[0029]

[0030] Where 0<β≤2, μ and v follow normal distribution;

[0031]

[0032] μ and v are random parameters that follow normal distribution in Levy's Fight algorithm;

[0033] in,

[0034]

[0035] β is used to control the tail behavior and jump frequency of the distribution; Г is the gamma function, which is used to adjust the shape of the probability density;

[0036] Round the obtained step size S and set the maximum step size S max , minimum step length S min S is modified, and Levy's Fight is applied to the walking behavior instead of the fixed step size S a .

[0037] Beneficial effects: The present invention improves the UAV task allocation scheme in many aspects and invents an improved method that can greatly improve the efficiency and correctness of UAV task allocation solution. It effectively solves the UAV task allocation problem under various constraints and provides intelligent and customizable intelligent task planning results for combat personnel. The following advantages are achieved:

[0038] (a) In a complex battlefield mission environment, a method that is independent of population initialization is urgently needed to eliminate the disadvantage of being highly dependent on population initialization and improve the accuracy of task allocation in a complex battlefield environment and multi-UAV multi-task scenario. The present invention uses chaos theory and reverse learning theory to optimize the initial distribution of the population, so that even a small number of populations can be better distributed in various locations in the solution space, eliminating the problems of insufficient population diversity and population aggregation that may be caused by random allocation, thereby eliminating the problem of high dependence on the population initialization effect in UAV mission planning.

[0039] (b) Multi-UAV multi-task allocation problem, even a single optimization target may be a multi-peak and multi-valley function. The objective function of multi-objective task planning is more complex, and the multi-peak and multi-valley phenomenon is more obvious. The fixed step size optimization strategy obviously cannot meet the needs of this task planning problem. The present invention uses a bionic algorithm existing in nature to simulate the step size characteristics of biological hunting that are ubiquitous in nature. A larger step size is used at long distances, and the step size is randomly increased when it may fall into a local optimal solution, thereby improving the ability to jump out of the local optimal solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is an accompanying drawing of the overall technical solution of the present invention.

[0041] Figure 2 is an example of a two-dimensional function. DETAILED DESCRIPTION

[0042] Example 1. A complex task planning method based on an improved wolf pack algorithm,

[0043] 1) Experimental environment:

[0044] The present invention implements the simulation environment of CRL-AMIWPA based on Matlab R2021b environment on Windows 11 operating system, and the PC configuration is: 12th Gen Intel(R)Core(TM)i7-12700H2.30GHz, 16G memory.

[0045] 2) Optimize target setting:

[0046] Suppose the time required to complete the reconnaissance mission is T zc , the time required to complete the attack task is T dj , for a single drone U i (i=1,2,3……N), let its task sequence be Among them, N zc Reconnaissance missions, N dj attack task, PreT k Represents the predecessor task number of task k, PreT k =0 means that this mission is the first mission of the UAV. represents the time taken from the predecessor task of k to task k, T O*k represents the time from the starting point to task k, then for a single UAV U i , the total time to complete the task is T Ui It can be expressed as:

[0047]

[0048] For the drone swarm, the time to complete the task is expressed as:

[0049]

[0050] Since the drone flies at a constant speed, its fuel consumption is proportional to the distance traveled. The longer the distance the drone flies, the greater the fuel consumption. Assuming the fuel consumption per unit distance is 1, represents the distance from the previous task of k to task k, D O，k represents the distance from the starting point to task k, then for a single UAV U i , the total distance to complete the task It can be expressed as

[0051]

[0052] The total distance D that the drone swarm travels to complete all tasks acc It is expressed as:

[0053]

[0054] In summary, the goal of task allocation is to minimize the completion time and fuel consumption. The cost function P is expressed as follows, where a and b are weighted coefficients used to balance the dimensions of time and distance and a+b=1:

[0055]

[0056] 3) Algorithm steps:

[0057] The overall technical solution of the present invention is shown in the attached Figure 1 shown.

[0058] Step 1, initialize the wolf pack algorithm parameters, including: wolf pack size N = 100, maximum number of iterations Maxgon = 100, wolf detection ratio factor α = 4, maximum number of wandering times S max =10, summoning distance threshold d near =3, wolf pack update ratio β = 0.4, using LF to generate walking step length sequence S a , Summoning step length S b =3, siege step length S c =1, Chaotic sequence X n , the initial value of the chaotic sequence X0 = 0.326789, LF generation process: β = 1.8, the step size is limited to [-0.3T num , 0.3T num ]. In addition, the UAV mission execution time is set as: 10 minutes for attack mission and 5 minutes for identification mission.

[0059] Step 2: Initialize the individual positions of wolves according to the chaotic sequence and reverse learning mechanism in step 1

[0060] Step 3: According to the individual objective function values ​​of the wolves, the objective values ​​are sorted and the best wolf is selected as the leader. The scout wolves of the wolf pack are selected according to the scout wolf ratio factor, and the remaining wolves become fierce wolves.

[0061] Step 4, determine whether the maximum number of iterations has been reached. If so, output the optimal solution, otherwise go to Step 5.

[0062] Step 5, the alpha wolf issues a wandering command to determine whether the scout wolf has reached the maximum wandering times. If so, go to Step 6, otherwise continue wandering. If the scout wolf's objective function value is better than the alpha wolf during wandering, the scout wolf replaces the alpha wolf as the alpha wolf of the wolf pack and immediately stops wandering, and go to Step 6.

[0063] Step 6: The alpha wolf issues a call, and the wolves in the pack respond to the call and rush toward the alpha wolf. It is determined whether the distance between the alpha wolf and the wolves is less than the distance threshold d. near , if less than d near If the objective function of the fierce wolf in the calling behavior is better than that of the alpha wolf, the fierce wolf replaces the alpha wolf and becomes the alpha wolf of the wolf pack. Then go to Step 7. Otherwise, continue the calling behavior until the distance with the alpha wolf is less than d. near .

[0064] Step 7: The alpha wolf issues a siege command, and all wolves except the alpha wolf respond to the command and carry out the siege.

[0065] Step 8, calculate the objective function value of the wolf pack, eliminate the worse individuals in the wolf pack according to the inferior wolf elimination mechanism, update the wolf pack and then go to Step 2.

[0066] Example 2. A complex task planning method based on an improved wolf pack algorithm, see Figure 1 ,

[0067] Step 1, initialization parameters, mainly initializing various parameters of the wolf pack algorithm, including: wolf pack size N, maximum number of iterations Maxgon, wolf exploration ratio factor, maximum number of wandering times, summoning distance threshold, wolf pack update ratio β, summoning step length, siege step length, where the summoning step length and siege step length decrease in sequence, simulating the characteristics of searching from long-distance large step length to close-distance small step length.

[0068] Step 2, chaos reverse learning initialization. The present invention uses Tent chaos mapping to evenly distribute the wolf pack in the entire solution space during initialization, and uses the idea of ​​reverse learning to further expand the wolf pack individuals and enrich the diversity of wolf pack individuals. Chaotic motion has the characteristics of randomness, regularity, and ergodicity. The equations that constitute the Tent sequence of the present invention are as follows:

[0069]

[0070] In the formula, x n is the element value of the nth generation tent sequence, x n+1 For the next generation, rand(0,1) is a random function;

[0071] Let the total number of tasks be T num , the total number of drones is U num , all possible individual wolf task allocation matrices from 1 to Numbered in sequence, starting with T num =2,U num=3 as an example, the task allocation matrix (1,1) is numbered 1, (1,2) is numbered 2, (2,1) is numbered 4, and (3,3) is numbered 9. Take the generated chaotic sequence value x i , the matrix number No represented by it is calculated as follows:

[0072]

[0073] Through the matrix number No, we can infer the matrix represented by No.

[0074] The concept of reverse learning was first used in the field of machine learning. In intelligent computing, the greater the diversity of the initial population, the more likely it is that a solution will be close to the global optimal solution. The ideal situation is that the global optimal solution is generated at the time of initialization, thereby improving the convergence speed of the algorithm and improving the search ability. The inventors have found that generating the opposite solution has a higher probability of finding the global optimum than generating a random solution. Therefore, on the basis of forming a chaotic random solution, the present invention further expands the population by obtaining the inverse solution of each individual based on the generated initial population. Suppose that the drone U i The task allocation matrix is ​​x i , the serial number of the drone assigned to task t is The total number of drones is U num , then for Its inverse value is:

[0075]

[0076] Step 3, calculate the objective function value of the individual wolves, sort the objective values, select the alpha wolf (optimal individual), and then select the scout wolf of the pack according to the scout wolf ratio factor, and the remaining wolves become fierce wolves. The objective function value is the weighted average of various indicators concerned after the drone allocation plan is selected (such as drone fuel consumption, distance, mission completion probability, revenue loss, etc.)

[0077] Step 4, determine whether the maximum number of iterations Maxgon is reached. If so, output the optimal solution, otherwise go to Step 5.

[0078] Step 5, the alpha wolf issues a wandering command to determine whether the exploratory wolf has reached the maximum wandering times. If it has reached the maximum wandering times, go to Step 6, otherwise it continues to wander. If the exploratory wolf's objective function value is better than the alpha wolf during the wandering process, the exploratory wolf replaces the alpha wolf as the alpha wolf of the wolf pack and immediately stops wandering, and goes to Step 6. Wandering is the intelligent behavior of the exploratory wolf. The exploratory wolf is an individual exploring in the wolf pack. Its step length is generated by the variable step length of the Levy's Fight algorithm to prevent the exploratory individual from falling into the local optimal solution and being unable to jump out, which affects the planning effect of the algorithm.

[0079] In a large space and a limited range, Levy's Fight is the best search strategy for local search, because it can ensure a fine search in a small range, and can jump out of the local area in a large range to avoid falling into the local optimal solution. The present invention uses the Mentegna algorithm to simulate the search process of Levy's Fight, and its step length calculation formula is:

[0080]

[0081] Where 0<β≤2, μ and v follow normal distribution

[0082]

[0083] Among them, μ and v are random parameters that obey the normal distribution in the Levy's Fight algorithm, whose values ​​obey the mathematical expectation of 0 and variances of σ respectively. μ and σ ν Normal distribution

[0084]

[0085] The parameter β usually takes a value in the range of 0<β<2 and controls the tail behavior and jump frequency of the distribution. The parameter Г is the gamma function. For positive integers n, Г(n) = (n+1)!, which is used to adjust the shape of the probability density and ensure that the parameters of the distribution achieve the desired mathematical properties.

[0086] In the present invention, the obtained step length S is rounded and the maximum step length S is set max , minimum step length S min S is modified to apply Levy's Fight to the wandering behavior instead of the fixed step length S a .

[0087] Step 6, the alpha wolf issues a call command, and the fierce wolves in the wolf pack respond to the command and rush towards the alpha wolf, and determine whether the distance between the alpha wolf and the fierce wolves is less than the distance threshold. If so, go to Step 7, otherwise continue the call behavior. If the fierce wolf's objective function is better than the alpha wolf in the call behavior, the fierce wolf replaces the alpha wolf as the alpha wolf of the wolf pack and goes to Step 7, otherwise continue the call behavior until the distance with the alpha wolf is less than the distance threshold.

[0088] Step 7: The alpha wolf issues a siege command, and all wolves except the alpha wolf respond to the command and carry out the siege.

[0089] Step 8, calculate the objective function value of the wolf pack, eliminate the worse individuals in the wolf pack according to the inferior wolf elimination mechanism, update the wolf pack and then go to Step 2.

Claims

1. A complex task planning method based on an improved wolf pack algorithm, characterized in that: The steps include: Step 1, initialize the wolf pack algorithm parameters; Step 2, perform chaos reverse learning initialization on the wolf pack initialized in Step 1; Step 3, calculate the objective function values ​​of the individual wolves in the wolf pack, and sort the objective values ​​to select the leader wolf, and then select the scout wolf of the wolf pack according to the scout wolf ratio factor, and the remaining wolves become fierce wolves; Step 4, determine whether the maximum number of iterations Maxgon is reached. If so, output the optimal solution, otherwise go to Step 5; Step 5, the alpha wolf issues a wandering command to determine whether the scout wolf has reached the maximum wandering times. If so, go to Step 6, otherwise continue wandering. If the scout wolf's objective function value is better than the alpha wolf during wandering, the scout wolf replaces the alpha wolf as the alpha wolf of the wolf pack and immediately stops wandering, and go to Step 6; Step 6, the alpha wolf issues a call command, and the fierce wolves in the wolf pack respond to the command and rush towards the alpha wolf, and determine whether the distance between the alpha wolf and the fierce wolves is less than the distance threshold. If so, go to Step 7, otherwise continue the call behavior; if the fierce wolf has a better objective function than the alpha wolf in the call behavior, the fierce wolf replaces the alpha wolf as the alpha wolf of the wolf pack and goes to Step 7, otherwise continue the call behavior until the distance with the alpha wolf is less than the distance threshold; Step 7: The alpha wolf issues a siege command, and all wolves except the alpha wolf respond to the command and engage in siege behavior; Step 8, calculate the objective function value of the wolf pack, eliminate the worse individuals in the wolf pack according to the inferior wolf elimination mechanism, update the wolf pack and then go to Step 2.

2. The complex task planning method based on the improved wolf pack algorithm according to claim 1 is characterized in that: In Step 1, the parameters of the wolf pack algorithm are initialized, including: wolf pack size N, maximum number of iterations Maxgon, wolf exploration ratio factor, maximum number of wandering times, summoning distance threshold, wolf pack update ratio β, summoning step length, and siege step length.

3. The complex task planning method based on the improved wolf pack algorithm according to claim 2 is characterized in that: The summoning step length and siege step length decrease successively.

4. The complex task planning method based on the improved wolf pack algorithm according to claim 1 is characterized in that: In Step 2, the tent chaotic mapping is used to evenly distribute the wolf pack in the entire solution space when the chaotic reverse learning is initialized. At the same time, the idea of ​​reverse learning is used to further expand the wolf individuals in the wolf pack and enrich the diversity of wolf individuals.

5. The complex task planning method based on the improved wolf pack algorithm according to claim 1 is characterized in that: In Step 2, the process of chaos reverse learning initialization is as follows: Construct the equation for the Tent sequence: In the formula, x n is the element value of the nth generation tent sequence, x n+1 For the next generation, rand(0,1) is a random function and the total number of tasks is T num , the total number of drones is U num , all possible individual wolf task allocation matrices from 1 to Number them in sequence; take the generated chaotic sequence value x i , the matrix number No represented by it is calculated as follows: Through the matrix number No, inversely deduce the matrix represented by No; UAV i The task allocation matrix is ​​x i , the serial number of the drone assigned to task t is The total number of drones is U num , then for Its inverse value is:

6. The complex task planning method based on the improved wolf pack algorithm according to claim 1 is characterized in that: In Step 3, the objective function value is the weighted average of various indicators of concern after the drone allocation scheme is selected.

7. The complex task planning method based on the improved wolf pack algorithm according to claim 1 is characterized in that: In Step 5, wandering is the intelligent behavior of the exploratory wolf, which is an individual exploring in the wolf pack. Its step length is generated by the variable step length of the Levy's Fight algorithm.

8. The complex task planning method based on the improved wolf pack algorithm according to claim 1 is characterized in that: In Step 5, the walking process is as follows: The Mentegna algorithm is used to simulate the search process of Levy's Fight. The step length calculation formula is: Where 0<β≤2, μ and v follow normal distribution; μ and v are random parameters that follow normal distribution in Levy's Fight algorithm; in, β is used to control the tail behavior and jump frequency of the distribution; Г is the gamma function, which is used to adjust the shape of the probability density; Round the obtained step size S and set the maximum step size S max , minimum step length S min S is modified to apply Levy's Fight to the wandering behavior instead of the fixed step length S a .

Images