A Multi-UAV Cooperative Reconnaissance Track Planning Method Based on Decision Maker Preferences

Abstract

This invention provides a multi-UAV cooperative reconnaissance trajectory planning method based on decision-maker preferences, comprising: Step 1, establishing a multi-objective optimization model for multi-UAV cooperative reconnaissance trajectory planning; Step 2, the decision-maker provides preference information, which includes the ideal value ranges of two optimization objectives, providing prior information for solving the multi-objective optimization model; Step 3, employing a preference-based multi-objective evolutionary algorithm, based on the prior information provided in Step 2 for solving the multi-objective optimization model, to optimize the multi-objective optimization model in Step 1, obtaining a multi-UAV cooperative reconnaissance trajectory planning flight scheme that satisfies the decision-maker's preferences. This invention can realize multi-UAV trajectory planning based on decision-maker preferences, ensuring the optimization performance of the solution, reducing computational resources, and alleviating the selection pressure on the decision-maker caused by too many candidate schemes.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) trajectory planning technology, specifically a multi-UAV collaborative reconnaissance trajectory planning method based on decision-maker preferences. Background Technology

[0002] Autonomous drone patrol systems offer advantages such as unmanned operation around the clock and across all areas, automatic alarms, and rapid response. They can be used for patrolling and surveillance of key locations such as ports, airports, and military bases. Multiple patrol drones carrying the same or different payloads can work collaboratively to achieve long-duration, all-weather, fully autonomous, and seamless patrols of the surveillance area, completing tasks such as routine patrols, rapid reconnaissance of key targets, and area-wide search. Drone trajectory planning involves planning the optimal or most satisfactory flight path for the drones, taking into account factors such as arrival time, power consumption, and flight environment, to ensure successful mission completion. Multi-drone collaborative reconnaissance trajectory planning not only requires planning a feasible optimal path for each drone but also considers the spatial and temporal coordination between them to achieve the best reconnaissance results.

[0003] Existing UAV trajectory planning technologies are mostly applied to reconnaissance in threatening environments, and the problem is often modeled as a single-objective optimization. Single-objective optimization models suffer from problems such as the inability to reasonably set weights and the difficulty in evaluating optimization performance. A few technologies consider multiple optimization objectives simultaneously, and the algorithm's result is a Pareto solution set for a multi-objective optimization problem, containing multiple flight paths. Selecting the final solution from this set requires the participation of a decision-maker. Generally, there are two approaches to solving multi-objective optimization problems: 1. A priori approach, where the decision-maker provides preference information, and the optimization algorithm finds a solution that satisfies the preference; 2. Posterior approach, where all Pareto solutions are first obtained, and then the decision-maker selects one. Most existing UAV trajectory planning problems modeled as multi-objective optimization problems adopt the posterior approach, i.e., obtaining the entire Pareto solution set first, and then the decision-maker selects one from it. As the problem size increases, obtaining the entire Pareto solution set becomes extremely difficult, and the optimization performance of the solution decreases accordingly. Furthermore, the Pareto solution set obtained by the posterior approach is enormous, making it difficult for the decision-maker to select one. Summary of the Invention

[0004] To address the aforementioned shortcomings of existing technologies, the purpose of this invention is to provide a multi-UAV collaborative reconnaissance trajectory planning method based on decision-maker preferences.

[0005] A multi-UAV cooperative reconnaissance trajectory planning method based on decision-maker preferences includes the following steps:

[0006] Step 1: Establish a multi-objective optimization model for collaborative reconnaissance trajectory planning using multiple UAVs;

[0007] Step 2: The decision-maker provides preference information, which includes the ideal value range of the two optimization objectives, providing prior information for solving the multi-objective optimization model;

[0008] Step 3: Using a preference-based multi-objective evolutionary algorithm, based on the prior information provided in Step 2 for solving the multi-objective optimization model, the multi-objective optimization model in Step 1 is solved and optimized to obtain a multi-UAV collaborative reconnaissance trajectory planning flight scheme that satisfies the decision-maker's preferences.

[0009] Furthermore, the method for establishing the multi-objective optimization model in step 1 is as follows:

[0010] Suppose there are N drones and M targets to be reconnoitered. The starting positions of the drones and the positions of the targets are known. The multi-drone collaborative reconnaissance trajectory planning is described as assigning the M targets to the N drones, determining the target access order of each drone, so that the total time for all targets to be reconnoitered is as short as possible, and the energy consumption of the drones is as low as possible.

[0011] Let the flight plan of UAV i be P. i ={UAV i ,T i1 ,T i2 ,…,T im}, where T ik Let k represent the reconnaissance target assigned to UAV i. Then, the total flight distance of UAV i is:

[0012]

[0013] The first objective function is the total time it takes for all targets to be detected. The goal is to make it as short as possible, where v is the drone's flight speed, and the second objective function is the total power consumption of all drones. As few as possible, of which e i Let be the power consumption per unit time of drone flight; then the multi-objective optimization model is expressed as:

[0014] minF(x) = [f1(x), f2(x)]

[0015] Where x represents the flight scheme of multiple drones, which must satisfy: (1) each drone must scout at least one target; (2) each target must be scout once; (3) the flight distance of each drone cannot exceed its power limit.

[0016] Furthermore, the preference information in step 2 is expressed as the ideal range of values ​​for the total time f1(x) for all targets being detected and the total power consumption f2(x) for all UAVs:

[0017]

[0018] Where L i and U i f i The lower and upper bounds of (x).

[0019] Furthermore, step 3 specifically includes:

[0020] Step 3.1: Initialize the population using a dual-chromosome encoding strategy: Chromosome 1 is the observation target sequence, and Chromosome 2 is the segmentation point of the observation target sequence. Chromosome 1 is divided according to the assigned UAVs.

[0021] Step 3.2 Generate weight vectors and define neighborhoods according to the classic MOEA / D algorithm, using preference information as the initial ideal point zi;

[0022] Step 3.3: Generate offspring individuals within the neighborhood using crossover and mutation operators: A partially matched crossover operator is used for population crossover. The mutation operation on chromosome 1 includes the following three mutation operators:

[0023] (1) The exchange operator randomly exchanges two genes in the original chromosome;

[0024] (2) The sliding operator randomly selects two gene loci on the original chromosome, moves the gene of the first gene loci to the second gene loci, and the sequence between the two gene loci slides forward in sequence.

[0025] (3) The reverse operator randomly selects two gene loci on the original chromosome and reverses the sequence between the two gene loci, while keeping other genes unchanged;

[0026] The probabilities of the crossover and mutation operators are controlled by the algorithm parameters;

[0027] Step 3.4, Evaluate Offspring Individuals: First, decode the offspring individuals according to the dual-chromosome encoding strategy, calculate the objective function values ​​f1(x) and f2(x) of the offspring individuals, and then assign fitness values ​​to the offspring individuals according to... Calculate the original fitness value, where z * Assuming a lower bound in the preference information, determine if the solution lies within the preference region. If so, transform the solution from the original target space to the preference region space and recalculate g. te (xλ i ,z * If the original fitness value is not found, a larger penalty value P is added to the original fitness value of the solution.

[0028] Step 3.5: Update the ideal point using the offspring population;

[0029] Step 3.6, Update the population: If solution x dominates solution y, then x will replace y regardless of the size of the fitness value; if x and y do not dominate each other, then x will replace y if and only if the fitness value of x is smaller.

[0030] Step 3.7: Determine whether the algorithm termination condition is met. If it is met, output the solution set for the decision-maker to check; if it is not met, return to step 3.3 to start a new round of evolutionary loop.

[0031] Step 3.8: The decision-maker checks the results calculated in step 3.7 and selects the final flight plan based on experience.

[0032] The beneficial effects of this invention are as follows: by using a priori methods based on decision-maker preference information to solve the UAV cooperative reconnaissance trajectory planning problem, compared with a posteriori methods, it can not only guarantee the optimization performance of the solution, but also reduce computational resources and alleviate the selection pressure on decision-makers caused by too many candidate solutions, thereby effectively solving the current problems of difficulty in solving the entire Pareto solution set and difficulty in decision-makers' selection for multi-UAV trajectory planning. Attached Figure Description

[0033] Figure 1 This is a flowchart illustrating a multi-UAV collaborative reconnaissance trajectory planning method based on decision-maker preferences, as described in an embodiment of the present invention.

[0034] Figure 2 This is a schematic diagram of the dual-chromosome encoding strategy in this invention;

[0035] Figure 3 This is a diagram illustrating the task scenario and planning results of an embodiment of the present invention. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] This invention considers the application of an autonomous unmanned aerial vehicle (UAV) patrol system in a scenario of rapid, coordinated reconnaissance of key targets. Under heightened alert conditions, rapid reconnaissance of multiple key targets (such as command posts and weapons depots) within a site is necessary to ensure their security. Upon receiving information about key targets (geographical location, importance level, etc.), the UAV system coordinates the reconnaissance of all targets as quickly as possible and transmits the results back to the command and control center. Flight path planning must balance timeliness and power consumption, prioritizing the planning scheme with high timeliness and low power consumption. However, there is a certain conflict between these two factors, requiring decision-makers to weigh the options based on specific circumstances to determine the final flight plan.

[0038] This invention is a multi-UAV cooperative reconnaissance trajectory planning method based on decision-maker preferences, and is implemented according to the following steps:

[0039] Step 1: Establish a multi-objective optimization model for collaborative reconnaissance trajectory planning using multiple UAVs;

[0040] In one embodiment, there are 3 UAVs and 62 targets to be reconnoitered. The starting positions of the UAVs and the positions (latitude and longitude coordinates) of the targets are known. Multi-UAV cooperative reconnaissance trajectory planning can be described as assigning the 62 targets to the 3 UAVs, determining the target access order (i.e., flight plan) for each UAV, such that the total time for all targets to be reconnoitered is as short as possible, and the energy consumption of the UAVs is as low as possible. Let the flight plan of UAV i be P. i ={UAV i ,T i1 ,T i2 ,…,T im Let i = 1, 2, 3, where Tik represents the k-th reconnaissance target assigned to UAV i, and there are m targets in total. Then the total flight distance of UAV i is...

[0041] The first objective function is the total time it takes for all targets to be detected. Where v is the flight speed of the UAV, and the second objective function is the total power consumption of all UAVs completing the reconnaissance mission. Where ei is the power consumption per unit time of UAV flight. The multi-objective optimization model can be expressed as minF(x)=[f1(x),f2(x)], where x represents the flight scheme of multiple UAVs, which must satisfy: (1) each UAV must scout at least 1 target; (2) each target must be scout once; (3) the flight distance of each UAV cannot exceed its power limit.

[0042] Step 2: The decision-maker provides preference information, namely the ideal value range of the two optimization objectives. In this example, the preferences are set according to the actual task requirements as follows:

[0043] 1800≤f1(x)≤2200,5100≤f2(x)≤6000.

[0044] Step 3: Optimize the model from Step 1 using a preference-based multi-objective evolutionary algorithm to obtain a multi-UAV cooperative reconnaissance trajectory planning flight scheme. The specific calculation process is illustrated using the T-MOEA / D multi-objective evolutionary algorithm based on decomposition and decision-maker preferences as an example.

[0045] Step 3.1: Initialize the population using a dual-chromosome coding strategy. Chromosome 1 is the target sequence, and chromosome 2 is the segmentation point of the target sequence. Chromosome 1 is divided according to the drones. For example, if chromosome 1 is [2 4 6 1 3 5] and chromosome 2 is [3 5], then chromosome 1 is cut after the 3rd and 5th genes, resulting in three subsequences [2 4 6], [1 3], and [5], which correspond to the flight schemes of the three drones, as shown below. Figure 2 As shown.

[0046] Step 3.2: Generate weight vectors and define neighborhoods according to the classic MOEA / D algorithm, and use preference information as the initial ideal point zi.

[0047] Step 3.3: Generate offspring individuals within the neighborhood using the crossover and mutation operator. The partial matching crossover operator (PMX) is used for population crossover. The mutation operation on chromosome 1 includes the following three mutation operators: (1) Exchange operator: randomly exchange two genes in the original chromosome; (2) Sliding operator: randomly select two gene positions on the original chromosome, move the gene from the first gene position to the second gene position, and slide the sequence between the two gene positions forward sequentially; (3) Reverse operator: randomly select two gene positions on the original chromosome, reverse the sequence between the two gene positions, and keep other genes unchanged. The probability of the crossover and mutation operator is controlled by the algorithm parameters.

[0048] Step 3.4: Evaluate offspring individuals. First, decode the offspring individuals according to the dual-chromosome encoding strategy, calculating the objective function values ​​f1(x) and f2(x) for each offspring. Then, assign fitness values ​​to the offspring individuals. Calculate the original fitness value, where z * This is the lower bound in the preference information. Determine if the solution lies within the preference region; if so, transform the solution from the original target space to the preference region space and recalculate g. te (xλ i ,z * If the original fitness value is not found, a larger penalty value P is added to the original fitness value of the solution.

[0049] Step 3.5: Update the ideal point using the offspring population.

[0050] Step 3.6: Update the population. In T-MOEA / D, Pareto dominance is considered a higher-order criterion than fitness value. That is, if solution x dominates solution y, then x will replace y regardless of its fitness value. If x and y have no dominance relationship, then x will replace y if and only if x has a smaller fitness value.

[0051] Step 3.7: Determine if the algorithm termination condition is met. If it is, output the solution set for the decision-maker to check; if not, return to step 3.3 to start a new round of evolutionary loop.

[0052] Step 3.8: The decision-maker checks the results calculated by the algorithm and selects the final flight plan. In this example, the decision-maker chooses the solution with f1(x) = 2049 and f2(x) = 5817 as the flight plan, as follows: Figure 3 As shown.

[0053] Experimental setup:

[0054] To verify the effectiveness of the proposed algorithm, the proposed algorithm T-MOEA / D and the posterior algorithm MOEA / D are compared. The experimental parameters are set as shown in Table 1:

[0055] Table 1 Simulation Experiment Algorithm Parameters

[0056]

[0057] The experiment set up two scenarios, representing two different preferences. Preference 1 has higher requirements for optimization objective 1 (total time), set as 1000≤f1(x)≤1400, 4400≤f2(x)≤5400. Preference 2 has higher requirements for optimization objective 2 (total power consumption), set as 1200≤f1(x)≤2000, 4000≤f2(x)≤4600. In both experimental scenarios, the T-MOEA / D algorithm and the MOEA / D algorithm were run 10 times each, calculating the Hypervolume index within the preference region. A larger Hypervolume index indicates better algorithm performance. The statistical data of the Hypervolume index obtained from the 10 calculations are shown in Table 2.

[0058] Table 2 Simulation Experiment Results

[0059]

[0060] The results show that the T-MOEA / D algorithm outperforms the MOEA / D algorithm in both scenarios in terms of Hypervolume metrics. Furthermore, the average runtime of the T-MOEA / D algorithm is 10.8 seconds, which is shorter than the MOEA / D algorithm's 23.5 seconds.

[0061] This invention is the first to apply a priori methods to UAV trajectory planning. Its advantages are: 1. It effectively solves the problems of difficulty and large computational load in solving the entire Pareto solution set by posterior methods, thus saving computational resources; 2. It considers preference information in the optimization process, resulting in fewer solutions and reducing the choice pressure on decision-makers.

[0062] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A multi-UAV cooperative reconnaissance trajectory planning method based on decision-maker preferences, characterized in that, Includes the following steps: Step 1: Establish a multi-objective optimization model for collaborative reconnaissance trajectory planning using multiple UAVs; Step 2: The decision-maker provides preference information, which includes the ideal value range of the two optimization objectives, providing prior information for solving the multi-objective optimization model; Step 3: Using a preference-based multi-objective evolutionary algorithm, based on the prior information provided in Step 2 for solving the multi-objective optimization model, the multi-objective optimization model in Step 1 is solved and optimized to obtain a multi-UAV cooperative reconnaissance trajectory planning flight scheme that satisfies the decision-maker's preferences. The method for establishing the multi-objective optimization model in step 1 is as follows: Suppose there are N drones and M targets to be reconnoitered. The starting positions of the drones and the positions of the targets are known. The multi-drone collaborative reconnaissance trajectory planning is described as assigning the M targets to the N drones, determining the target access order of each drone, so that the total time for all targets to be reconnoitered is as short as possible, and the energy consumption of the drones is as low as possible. Let the flight plan of UAV i be... T im Let m represent the m-th reconnaissance target assigned to UAV i. Then, the total flight distance of UAV i is: ； The first objective function is the total time it takes for all targets to be detected. The goal is to make it as short as possible, where v is the drone's flight speed, and the second objective function is the total power consumption of all drones. As few as possible, of which e i Let be the power consumption per unit time of drone flight; then the multi-objective optimization model is expressed as: ； Where x represents the flight scheme of multiple drones, which must satisfy: (1) each drone must scout at least one target; (2) each target must be scout once; (3) the flight distance of each drone cannot exceed its power limit.

2. The method according to claim 1, characterized in that, The preference information in step 2 is expressed as the total time all targets were detected. Total power consumption of all drones Ideal range of values: ； in and f i The lower and upper bounds of (x).

3. The method according to claim 1, characterized in that, Step 3 specifically includes: Step 3.1: Initialize the population using a dual-chromosome encoding strategy: Chromosome 1 is the observation target sequence, and Chromosome 2 is the segmentation point of the observation target sequence. Chromosome 1 is divided according to the assigned UAVs. Step 3.2 Generate weight vectors and define neighborhoods according to the classic MOEA / D algorithm, using preference information as the initial ideal point zi; Step 3.3: Generate offspring individuals within the neighborhood using crossover and mutation operators: A partially matched crossover operator is used for population crossover. The mutation operation on chromosome 1 includes the following three mutation operators: (1) The exchange operator randomly exchanges two genes in the original chromosome; (2) The sliding operator randomly selects two gene loci on the original chromosome, moves the gene of the first gene loci to the second gene loci, and the sequence between the two gene loci slides forward in sequence. (3) The reverse operator randomly selects two gene loci on the original chromosome and reverses the sequence between the two gene loci, while keeping other genes unchanged; The probabilities of the crossover and mutation operators are controlled by the algorithm parameters; Step 3.4, Evaluate offspring individuals: First, decode the offspring individuals according to the dual-chromosome encoding strategy and calculate the objective function value of the offspring individuals. and Then, fitness values ​​are assigned to the offspring individuals, based on... Calculate the original fitness value, where Using the lower bound of the preference information, determine whether the offspring individual is located within the preference region. If so, transform the offspring individual from the original target space to the preference region space and recalculate. If the original fitness value is not a new fitness value, then a larger penalty value P is added to the original fitness value of the offspring. Step 3.5: Update the ideal point using the offspring population; Step 3.6, Update the population: If solution x dominates solution y, then x will replace y regardless of the size of the fitness value; if x and y do not dominate each other, then x will replace y if and only if the fitness value of x is smaller. Step 3.7: Determine if the algorithm termination condition is met. If it is met, output the solution set for the decision-maker to check; if it is not met, return to step 3.3 to start a new round of evolutionary loop. Step 3.8: The decision-maker checks the results calculated in step 3.7 and selects the final flight plan based on experience.

Images