A multi-party multi-objective optimization method for unmanned aerial vehicle path planning and related equipment

Abstract

The embodiment of the present application provides a multi-party multi-objective optimization method for unmanned aerial vehicle path planning, including: obtaining an initial population corresponding to a multi-objective decision function of a planned path of the unmanned aerial vehicle, and performing non-dominated sorting on the individuals to be optimized in the initial population to obtain the non-dominated levels of the individuals to be optimized; sorting according to the non-dominated levels, activating the individuals to be optimized with higher non-dominated sorting using a multi-party adaptive activation operator to obtain a set of activated optimized individuals; calculating the crowding distance information of the activated optimized individuals, and obtaining a set of cloned optimized individuals according to the crowding distance information and the non-dominated levels; generating offspring optimized individuals according to the cloned optimized individuals using a multi-party adaptive crossover operator algorithm; merging the offspring optimized individuals with the initial population to obtain a candidate population; sorting the candidate population according to the selection operator algorithm and the crowding distance information, and selecting the target optimized individuals to participate in the cycle until the maximum number of cycles is reached as the optimization scheme of the target decision function.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of drone path planning, and in particular to a multi-party and multi-objective optimization method for drone path planning and related equipment. Background Art

[0002] Multi-party multi-objective optimization problem (MPMOP) is a special type of multi-objective optimization problem. It describes a scenario with multiple optimization objectives and multiple decision makers participating in the decision-making. Each decision maker often cares about multiple objectives, and these objectives are also conflicting. A typical problem is the unmanned aerial vehicle path planning problem (UAVPP). For example, drone users usually require short paths and low energy consumption. Safety decision makers usually require that the path poses little third-party risk to pedestrians and vehicles, and has little noise pollution, etc. Using multi-party multi-objective optimization problems to characterize the path planning problem of drones is more in line with the needs than general multi-objective optimization problems. Therefore, the construction and testing of multi-party multi-objective drone path planning problems is a meaningful study.

[0003] Designing an algorithm to solve multi-party multi-objective optimization problems requires considering the interests of each decision maker. However, using existing multi-objective evolutionary optimization algorithms to solve multi-party multi-objective optimization problems will face problems such as non-convergence of solution sets and lack of diversity. Therefore, ordinary multi-objective evolutionary optimization algorithms are not suitable for multi-party multi-objective optimization problems. Currently, there are few studies on existing multi-party multi-objective optimization algorithms, and they have not been studied in depth.

[0004] Therefore, how to construct and design an efficient multi-party and multi-objective evolutionary optimization algorithm to better solve multi-party and multi-objective optimization problems is a technical problem that needs to be solved urgently. Summary of the invention

[0005] The embodiments of the present application provide a multi-party multi-objective optimization method and related equipment for unmanned aerial vehicle path planning, which are used to solve multi-party multi-objective optimization problems so that most individuals in the population can provide information.

[0006] A first aspect of an embodiment of the present application provides a multi-party and multi-objective optimization method for drone path planning, comprising:

[0007] Obtaining an initial population in the objective function space corresponding to the planned path of the UAV of the multi-objective decision function, and performing non-dominated sorting on each individual to be optimized in the initial population to obtain the non-dominated level of all individuals to be optimized; wherein the multi-objective decision function includes an efficiency decision function and a safety decision function, the objective function space includes a function matrix of the planned path of the UAV from the perspective of the efficiency decision function, and a function matrix of the planned path of the UAV from the perspective of the safety decision function, and the initial population includes initial function solutions generated by the efficiency decision function and the safety decision function in the objective function space;

[0008] When the current evaluation times of any objective decision function do not meet the preset evaluation times, the non-dominated levels are sorted, and the individuals to be optimized with the top non-dominated rankings are activated according to the multi-party adaptive activation operator algorithm to obtain an activated optimized individual set; wherein the activated optimized individual set includes the activated optimized individual, and the activated optimized individual is used to characterize the individuals to be optimized after the adaptive activation;

[0009] According to the clonal operator algorithm, the crowding distance information of each activated optimized individual in the activated optimized individual set is calculated, so that in the activated optimized individual set, resources are allocated according to the crowding distance information and the non-dominated level corresponding to each activated optimized individual to obtain a clonal optimized individual set; wherein the clonal optimized individual set includes clonal optimized individuals, and the clonal optimized individuals are used to characterize the activated optimized individuals after resource allocation;

[0010] According to a multi-party adaptive crossover operator algorithm, based on all clone optimized individuals in the clone optimized individual set, offspring optimized individuals are generated;

[0011] Merging the offspring optimized individuals with the initial population to obtain a candidate population;

[0012] The candidate population is sorted according to the selection operator algorithm and the crowding distance information, and the target optimized individual is selected from the sorted candidate population as the population participating in the next round of cycles until the maximum number of cycles is reached, and the target optimized individual that meets the maximum number of cycles is used as the optimization solution of the target decision function; wherein the target optimized individual that reaches the maximum number of cycles is the target function solution of the efficiency decision function and the safety decision function corresponding to the planned path of the drone in the target function space.

[0013] A second aspect of the embodiment of the present application provides a multi-party multi-objective optimization device, including:

[0014] An acquisition unit is used to acquire an initial population in an objective function space corresponding to a multi-objective decision function of a planned path of a UAV, and to perform non-dominated sorting on each individual to be optimized in the initial population to obtain a non-dominated level of all individuals to be optimized; wherein the multi-objective decision function includes an efficiency decision function and a safety decision function, the objective function space includes a function matrix of the planned path of the UAV from the perspective of the efficiency decision function, and a function matrix of the planned path of the UAV from the perspective of the safety decision function, and the initial population includes an initial function solution generated by the efficiency decision function and the safety decision function in the objective function space;

[0015] A sorting unit is used for sorting according to the non-dominated level when the current evaluation times of any objective decision function do not meet the preset evaluation times, and adaptively activating the individuals to be optimized with the top non-dominated ranking to obtain an activated optimized individual set; wherein the activated optimized individual set includes the activated optimized individuals, and the activated optimized individuals are used to characterize the individuals to be optimized after the adaptive activation;

[0016] A calculation unit, configured to calculate the crowding distance information of each activated optimization individual in the activated optimization individual set according to the clonal operator algorithm, so as to allocate resources in the activated optimization individual set according to the crowding distance information and the non-dominated level corresponding to each activated optimization individual, and obtain a clonal optimization individual set; wherein the clonal optimization individual set includes clonal optimization individuals, and the clonal optimization individuals are used to characterize the activated optimization individuals after resource allocation;

[0017] A generating unit, configured to generate offspring optimized individuals according to all clone optimized individuals in the clone optimized individual set according to a multi-party adaptive crossover operator algorithm;

[0018] A merging unit, used for merging the offspring optimized individuals with the initial population to obtain a candidate population;

[0019] A selection unit is used to sort the candidate population according to the selection operator algorithm and the crowding distance information, and select the target optimized individual from the sorted candidate population as the population participating in the next round of cycles until the maximum number of cycles is reached, and the target optimized individual that meets the maximum number of cycles is used as the optimization solution of the target decision function; wherein the target optimized individual that reaches the maximum number of cycles is the target function solution of the efficiency decision function and the safety decision function corresponding to the planned path of the drone in the target function space.

[0020] The multi-party multi-objective optimization device provided in the second aspect of the embodiment of the present application is used to execute the multi-party multi-objective optimization method for drone path planning described in the first aspect.

[0021] A third aspect of the embodiments of the present application provides a multi-party multi-objective optimization device, including:

[0022] CPU, memory, input and output interfaces, wired or wireless network interfaces, and power supply;

[0023] The memory is a short-term storage memory or a persistent storage memory;

[0024] The central processing unit is configured to communicate with the memory and execute instruction operations in the memory to perform the multi-party and multi-objective optimization method for drone path planning described in the first aspect.

[0025] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium, which includes instructions. When the instructions are executed on a computer, the computer executes the multi-party and multi-objective optimization method for drone path planning described in the first aspect.

[0026] A fifth aspect of an embodiment of the present application provides a computer program product, which includes instructions. When the instructions are executed on a computer, the computer executes the multi-party and multi-objective optimization method for drone path planning described in the first aspect.

[0027] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages: through a multi-party multi-objective optimization method for drone path planning disclosed in the embodiments of the present application, an efficient optimization algorithm capable of solving multiple objectives containing multiple decision-makers (objective decision functions) is designed. Among them, providing guidance from the perspective of an objective decision function leads to a significant increase in the number of individuals in the population that can provide information. Therefore, in the multi-party multi-objective optimization problem, according to this strategy, most individuals in the population may be able to provide guidance information. Therefore, compared with the traditional multi-objective optimization algorithm, it has the characteristics of high algorithm solving performance and rich diversity of solution schemes. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0029] Figure 1 A schematic diagram of a drone path planning target, constraint, and decision-making method disclosed in an embodiment of the present application;

[0030] Figure 2 A flowchart of a multi-party and multi-objective optimization method for UAV path planning disclosed in an embodiment of the present application;

[0031] Figure 3 A flowchart of another multi-party multi-objective optimization method for UAV path planning disclosed in an embodiment of the present application;

[0032] Figure 4 A flowchart of another multi-party multi-objective optimization method for UAV path planning disclosed in an embodiment of the present application;

[0033] Figure 5 A flowchart of another multi-party multi-objective optimization method for UAV path planning disclosed in an embodiment of the present application;

[0034] Figure 6 A flowchart of another multi-party multi-objective optimization method for UAV path planning disclosed in an embodiment of the present application;

[0035] Figure 7 A schematic diagram of a multi-party guided crossover strategy in a multi-party adaptive crossover operator disclosed in an embodiment of the present application;

[0036] Figure 8 A schematic diagram of an existing activation method (existing method) and a multi-party adaptive activation operation disclosed in an embodiment of the present application;

[0037] Fig. 9 An overall framework diagram of an MPIA disclosed in an embodiment of the present application;

[0038] Fig.10 A schematic diagram of the structure of a multi-party multi-objective optimization device disclosed in an embodiment of the present application;

[0039] Fig.11 A schematic diagram of the structure of a multi-party multi-objective optimization device disclosed in an embodiment of the present application. DETAILED DESCRIPTION

[0040] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0041] It should be noted that the descriptions involving "first", "second", etc. in this application are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in this field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0042] In traditional multi-objective optimization problems / UAV path optimization problems, there is usually only one decision maker (DM), and the boundary individuals on its Pareto front are usually part of the Pareto non-dominated set. However, in multi-party multi-objective optimization problems (MPMOPs) / UAV path optimization problems, this is not always the case. Therefore, the strategy of fixed activation of head non-dominated individuals in traditional immune algorithms may result in the problem of lack of diversity in the final solution set. However, this strategy is used in most immune algorithms. In this scenario, a more flexible strategy is needed to dynamically select and activate individuals to ensure that the population can effectively explore and utilize the diversity and convergence in the multi-party objective space. Therefore, the present invention designs an adaptive activation strategy that can dynamically adjust the number of activated individuals and the selection criteria according to the state of the population and the characteristics of the multi-party objective space. This strategy aims to maximize the diversity and convergence of the population while ensuring that the algorithm can effectively explore and utilize potential solutions in the multi-party objective space. In addition, using this approach in solving multi-party multi-objective optimization problems (MPMOPs) has the advantage of being more intuitive than using it in solving uni-party multi-objective optimization problems (MOPs). In MOPs involving a single decision maker (DM), the guidance information is usually provided by a few individuals near the frontier. Therefore, the population converges to these guiding individuals. Without proper policy control, the population may face potential local convergence and lack of diversity.

[0043] In view of the above technical difficulties, the algorithm designed by the technical solution of this application to solve the multi-party multi-objective optimization problem needs to consider the interests of each decision maker. When using the existing multi-objective evolutionary optimization algorithm to solve the multi-party multi-objective optimization problem, problems such as non-convergence of the solution set and lack of diversity will be faced. Therefore, ordinary multi-objective evolutionary optimization algorithms are not suitable for multi-party multi-objective optimization problems. At present, there are few studies on the existing multi-party multi-objective optimization algorithms, and they have not been studied in depth. The technical solution of this application systematically studies the multi-party multi-objective unmanned aerial vehicle path planning problem (MPMOUAVPP, Multi-Party Multi-Objective Unmanned Aerial Vehicle Path PlanningProblem) with multiple decision-makers, constructs a series of unmanned aerial vehicle path planning problems with multiple decision-makers, and expands some existing multi-objective evolutionary optimization algorithms into multi-party multi-objective optimization algorithms that can solve this problem. At the same time, an efficient multi-party multi-objective evolutionary optimization algorithm is designed according to the characteristics of the problem to better solve the multi-party multi-objective optimization problem, and it is applied to the multi-party multi-objective unmanned aerial vehicle path planning problem, and compared with the solution effect of other existing algorithms. Furthermore, the multi-objective algorithm is to find a set of mutually Pareto non-dominated solutions. To put it simply, for example, I have three cars, one is fuel-efficient but slow, one is fuel-consuming but fast, and one is neither fuel-efficient nor fast. Of these three cars, the first two are the solutions we care about, and the third is dominated by the first two. The multi-objective algorithm is to find as many Pareto non-dominated solutions as possible. This article considers a multi-party multi-objective algorithm, which is a solution set with two decision parties. For example, for A, whether the car is fast and fuel-efficient, and for B, whether it is good-looking and safe. This set of solutions is required to meet the Pareto non-dominated requirements from A's point of view, and also meet the non-dominated requirements of B's ​​goals.

[0044] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0045] For easier understanding, please refer to Figure 1 , Figure 1 The following is a schematic diagram of the objectives, constraints, and decision-making methods of a drone path planning disclosed in an embodiment of the present application. Figure 1It can be seen that in the path planning problem of drones, path optimization based on the perspective of efficiency maximization is the most intuitive and common decision-making requirement. Here, this type of decision-making goal is summarized as the efficiency decision-maker, and based on the track segment of the drone's planned path, how to calculate these objective functions is discussed from the perspective of the efficiency decision-maker. For the efficiency decision-maker, maximizing the flight efficiency of the drone is the key requirement of the efficiency decision-maker. This requirement is mainly reflected in the goals of minimizing the length of the drone's flight trajectory, minimizing the change in flight altitude, minimizing flight energy consumption, and minimizing the sum of the distances of the mission hovering points. In the process of drone missions, the length of the drone's planned path directly affects the drone's operating time. Users who focus on efficiency often want to choose the shortest path to shorten the operating time. The safety decision-maker makes the drone risk controllable by optimizing these goals, including reducing the fatal risk of pedestrians and moving vehicles, reducing the property risk of buildings, and reducing noise pollution in urban areas. Therefore, in order to comprehensively consider the goals of efficiency and safety, as well as the constraints related to drone performance, terrain restrictions, etc., it is necessary to combine objective functions such as decision efficiency or safety decision, and combine the altitude, pitch angle or turning angle of the drone during flight, and various factors such as path length (a), drone energy consumption (b) (see Figure 1 ),,,,,noise pollution (h) and other constraints to find the optimal solution under multiple objectives.

[0046] Therefore, see Figure 7 , Figure 8 and Fig. 9 .in, Figure 7 A schematic diagram of a multi-party guided cross strategy disclosed in an embodiment of the present application is shown in FIG. Figure 8 This is a schematic diagram of an inappropriate activation operation disclosed in an embodiment of the present application. Fig. 9 The overall framework diagram of a multi-party immune algorithm (MPIA) disclosed in an embodiment of the present application.

[0047] Depend on Fig. 9 It can be seen that the MPIA proposed in the technical solution of this application can be used to solve the multi-party and multi-target UAV path planning problem. Specifically, Fig. 9 The main process of an iteration in MPIA is shown, such as Fig. 9As shown. The first subfigure describes the population distribution in the target space from the perspective of different decision makers (DMs). Operation (a), i.e., Multi-Party Non-dominated Sorting (MPNDS2), sorts the population to obtain a multi-party multi-objective non-dominated rank. Operation (b), i.e., adaptive activation, is used to select excellent individuals for two DMs as parent individuals for generating offspring. Operation (c), i.e., cloning operation, uses non-dominated sorting rank information and crowding distance information to allocate replication resources. Operation (d), i.e., the multi-party adaptive crossover operator algorithm is an adaptive operation based on multiple differential algorithms (DE, Differential Evolution Algorithm) operators and multi-party guided crossover operators. Operation (e), i.e., the selection operator algorithm uses multi-party non-dominated sorting and crowding distance to select the population to enter the next round of iteration. Among them, the horizontal axis is f1 and the vertical axis is f2 from the perspective of DM1; the horizontal axis is f3 and the vertical axis is f4 from the perspective of DM2. For operation (d.1), the operator used may be the rand / 2 / bin DE operator (CR=0.9, F=0.7), and the other operator may be the rand / 1 / bin DE operator (CR=0.5, F=0.5). For details, please refer to the embodiments shown later.

[0048] Furthermore, for operation (d), see Figure 7 , Figure 7 Specifically, it is a crossover operator based on a multi-party guided crossover strategy. This strategy intelligently selects guided individuals based on their non-dominated rankings obtained from different DM perspectives. Suppose that we consider an MPMOP with two decision makers (i.e., DM1 and DM2), each of which considers two different decision objectives, for a total of four objectives. Figure 7 (a) shows the population from the perspective of DM1. Figure 7 (b) in the figure shows the population from the perspective of DM2. The hexagonal individuals will serve as guided individuals, while the circular individuals will serve as guiding individuals. This strategy aims to search for new individuals that are located on the PF of the current two DMs. The figure illustrates the basic concept of the proposed multi-party guided crossover. Using this method when solving MPMOPs has the advantage of being more intuitive than using it when solving MOPs. In MOPs involving a single DM, guidance information is usually provided by a few individuals near the frontier. Therefore, the population converges to these guiding individuals. Without proper strategy control, the population may experience potential local convergence and lack diversity. However, in multi-party multi-objective optimization problems, guidance only needs to be provided from the perspective of one DM, resulting in a significant increase in the number of individuals that can provide guidance information. Therefore, in multi-party multi-objective optimization problems, according to this strategy, most individuals in the population may be able to provide guidance information.

[0049] Furthermore, for operation (c), see Figure 8 In MPMOPs, individuals on the Pareto front (PF) boundary of a single DM are not always part of the Pareto non-dominated set. For example, consider five solutions ,exist Fig. 9 In the figure, individuals marked in black constitutes a multi-party common Pareto set (PS). Although other solutions are not part of the common PS, they may still be individuals on the PF or boundary of a single DM. This is because the algorithm uses a multi-party non-dominated sorting operation, such as MPNDS or MPNDS2, which makes it possible that some individuals on the boundary of the target space are not in the multi-party Pareto set. Specifically, existing MPMOEAs directly select a fixed number of activated individuals or select individuals from the common Pareto frontier. Therefore, these algorithms cannot maintain the diversity of offspring in the target space of different DMs. They also have difficulty in effectively using individuals located on the PF of the target space of a specific DM to generate offspring, although these individuals contain valuable information. When individuals are activated using the existing multi-party immune algorithm, although most of these individuals are located on the common PF of the two DMs, they fail to represent the population information of DM1. In contrast, using a multi-party adaptive activation operator algorithm for activation operations can more ideally activate individuals, where the activated individuals cover the frontiers of different DMs.

[0050] To understand the above Fig. 9 For technical details, see Figure 2 , Figure 2 The flowchart of a multi-party multi-objective optimization method for UAV path planning disclosed in an embodiment of the present application includes steps 201 to 206.

[0051] 201. Obtain an initial population in the objective function space of the multi-objective decision function corresponding to the UAV planning path, and perform non-dominated sorting on each individual to be optimized in the initial population to obtain the non-dominated level of all individuals to be optimized.

[0052] In order to solve the MPMOPs problem, it is necessary to obtain the initial population in the objective function space under multiple objective decision functions, so as to perform non-dominated sorting on each individual to be optimized in the initial population, thereby obtaining the non-dominated level of all individuals to be optimized.

[0053] In one specific embodiment, the target decision function corresponds to Fig. 9In the above, there are different decision-makers (such as efficiency decision-makers or safety decision-makers), and the corresponding decision functions are efficiency decision functions and safety decision functions, which are not described in detail here. Furthermore, the objective function space is the function matrix from the perspective of different decision-makers (DM1 or DM2). For details, please refer to Fig. 9 The first subgraph in . Therefore, the initial population can be understood as the target conditions considered in the above-mentioned path length, drone energy consumption or path height change. For example, the initial population, that is, the random initialization scheme, uses uniformly distributed random numbers to generate corresponding individuals in the upper and lower bounds of the decision variable, and evaluates these individuals. For example, when the population size is 100 in the drone problem solution, the initialization is to randomly generate 100 drone path solutions (the path solution can be discretized into path points. If each solution has 50 points, then it is a vector of 150 length. In other words, different drone path solutions are individuals to be optimized), that is, to generate a 100x150 random matrix. Furthermore, the initial population can be directly randomly generated, or a general solution can be manually designed. Therefore, the target conditions that need to be considered, such as the path length, drone energy consumption or path height change, can be obtained through drone aerial photography or other public data of other enterprises.

[0054] Therefore, for different initial populations, it is necessary to use the multi-party non-dominated sorting operator MPNDS2. It is not difficult to understand that MPNDS2 regards the number of non-dominated layers under different decision makers (DMs) as the target, and uses two rounds of non-dominated ranking to calculate the non-dominated layers of the multi-objective optimization problem MPMOPs. It should be noted that in the non-dominated sorting process, the corresponding non-dominated sorting level will be generated through the sorting results. For easy understanding, please refer to Figure 3 The embodiment shown.

[0055] Furthermore, before performing non-dominated sorting on the initial population, the target value of the initial population is calculated for different DMs. The target value is calculated by calculating various objective functions that we need to consider based on the path, such as path length and noise. The calculation process and method of the target value are not specifically restricted here.

[0056] Furthermore, in other feasible technical solutions, in addition to the multi-objective decision function, any decision function can be used, such as efficiency decision function, safety decision function, cost decision function, etc., without specific scenario requirements. This embodiment does not limit this, and in the optimization scheme, the multi-objective decision function is preferably selected as the specific solution of this embodiment.

[0057] 202. When the current evaluation times of any objective decision function does not meet the preset evaluation times, the individuals to be optimized with the highest non-dominated ranking are sorted according to the non-dominated ranking, and the individuals to be optimized with the highest non-dominated ranking are activated according to the multi-party adaptive activation operator algorithm to obtain the activated optimized individual set.

[0058] Therefore, after completing the non-dominated sorting of the initial population and solving the MPMOPs, when the current evaluation times of any objective decision function do not meet the preset evaluation times, they are sorted according to the non-dominated level, so that the individuals to be optimized with the top non-dominated sorting are adaptively activated according to the multi-party adaptive activation operator algorithm to obtain the activated optimized individual set. It should be noted that the activated optimized individual set includes the activated optimized individual, which is used to characterize the individuals to be optimized after the adaptive activation.

[0059] In one of the specific embodiments, the individual to be optimized with the largest non-dominated level in the initial population is first determined in advance, and then the maximum number of evaluations of the individual to be optimized with the largest non-dominated level is set, wherein the number of evaluations in this embodiment can be understood as the number of iterations. That is to say, when the number of iterations of the population corresponding to the individual to be optimized with the largest non-dominated level does not reach the maximum number of evaluations, it can be sorted in sequence according to the non-dominated level, that is, rank. Among them, there is no restriction on the sorting method. However, in this embodiment, it is usually sorted from small to large. Then, in the process of sorting the non-dominated level, the individuals to be optimized that are at the front (for rank, it is from small to large, and the lowest rank is retained) are adaptively activated to obtain a set of activated optimized individuals.

[0060] Furthermore, a certain proportion of high-quality individuals are selected from the population for cloning and improved individuals are generated. The performance of different sets of activated individuals is evaluated using the value of the multi-party coverage metric (MCM). Specifically, MCM measures the proportion of the Pareto front distribution range of a given set of activated individuals from the perspective of the decision maker in the worst case. The higher the MCM value, the better the diversity of the given set of activated individuals among all decision makers. For ease of understanding, please refer to Figure 4 The embodiment shown.

[0061] 203. According to the clonal operator algorithm, the crowding distance information of each activated optimization individual in the activated optimization individual set is calculated, and in the activated optimization individual set, resources are allocated according to the crowding distance information and the non-dominated level corresponding to each activated optimization individual to obtain the clonal optimization individual set.

[0062] Then, the crowding distance information of each activated optimization individual in the activated optimization individual set can be calculated according to the clonal operator algorithm, so that in the activated optimization individual set, resources are allocated according to the crowding distance information and the non-dominated level corresponding to each activated optimization individual to obtain a cloned optimization individual set. It should be noted that the cloned optimization individual set includes cloned optimization individuals, and the cloned optimization individuals are used to characterize the activated optimization individuals after resource allocation.

[0063] In one of the specific embodiments, more replication resources are allocated to activated individuals with higher quality to increase their chances of cross-mutation and producing offspring. However, considering the adaptive activation strategy proposed in the technical solution of the present application, some activated individuals may not be the Pareto solution of MPMOP. Therefore, it is crucial to keep the population convergent. In order to achieve this, the technical solution of the present application uses crowding distance and combines convergence metrics to effectively allocate computing resources during the cloning process. Therefore, it is necessary to obtain the crowding distance of each activated optimized individual. Specifically, the crowding distance is used to measure the diversity of solutions. For an individual, the nearest other individuals in the objective function space are the crowding distance of this individual. If the crowding distance is larger, it means that the individual is more precious, because no other individual can replace it. Therefore, when retaining individuals, they will be retained from small to large according to rank until the individuals of a certain rank level are retained when they exceed the population size. Then, some individuals are selected by crowding distance. At the same time, the number of multi-party non-dominated layers is also used as one of the indicators of resource allocation. Therefore, through the comprehensive consideration of non-dominated levels and crowding information, activated optimized individuals can be obtained, thereby obtaining an activated optimized individual set. Therefore, resources can be allocated to each activated optimized individual through multiple non-dominated sorting layers and crowding distance information to obtain cloned optimized individuals and a set of cloned optimized individuals.

[0064] For easier understanding, please refer to Figure 5 The embodiment shown.

[0065] 204. According to the multi-party adaptive crossover operator algorithm, all clone optimized individuals in the clone optimized individual set are used to generate offspring optimized individuals.

[0066] Based on step 203, a multi-party adaptive crossover operator algorithm can be selected to generate offspring optimized individuals according to all clone optimized individuals in the clone optimized individual set.

[0067] In one specific embodiment, for all clone optimization individuals in the clone optimization individual set, a differential evolution (DE) operator and the cross-guidance strategy are proposed. Specifically, a set of rules are designed to ensure that these operators are appropriately applied during the algorithm process, wherein the rule is an adaptive operation operator algorithm. For ease of understanding, please refer to Figure 5 The embodiment shown.

[0068] Furthermore, in this embodiment, generation of offspring optimized individuals can be achieved by cloning optimized individuals and correspondingly activating optimized individuals.

[0069] 205. Merge the offspring optimized individuals with the initial population to obtain a candidate population.

[0070] Based on step 205, the offspring optimized individuals can be merged with the initial population to obtain a candidate population.

[0071] In one specific embodiment, in step 204, each cloned optimized individual and the corresponding activated optimized individual can generate a descendant optimized individual. At the same time, the initial population should also be understood as the initial population after iteration (the first iteration is the most original population). After cross-merging all descendant optimized individuals with the initial population, the polynomial mutation (PM) operator is used to perform mutation operations on the individuals to obtain a candidate population.

[0072] 206. Sort the candidate population according to the selection operator algorithm and the crowding distance information, and select the target optimized individual from the sorted candidate population as the population participating in the next round of cycles until the maximum number of cycles is reached, and use the target optimized individual that meets the maximum number of cycles as the optimization solution of the target decision function.

[0073] In one specific embodiment, solutions are retained layer by layer according to the number of layers of the multi-party non-dominated sorting operator MPNDS2. In a given layer, if the total number of retained solutions exceeds the number of maintained populations, the crowding distance information will filter out some crowded solutions in the layer, which is conducive to maintaining the diversity of the population. Therefore, by removing the most crowded individuals in the non-dominated sorting layer from the population sorted by the MPNDS2 operator, the target optimized individual can be obtained. Thus, the solution of the final target decision function (DM) of the optimized individual is obtained. For ease of understanding, please refer to Figure 6 The embodiment shown.

[0074] Combine steps 201 to 206, and Fig. 9 From the description shown, it can be seen that this embodiment mainly uses MPIA to solve the multi-party multi-objective problem MPMOPS. To facilitate understanding of the description, a supplementary description is given in pseudo code form. Please refer to the following statement:

[0075] It should be noted that FE is the number of feasible evaluations, which is used to indicate the number of evaluations of the preset maximum individual (objective function). F0 refers to the objective function matrix of the initial population. E0 is the path length or noise described above. For ease of understanding, this will not be described in detail later. Furthermore, in the first round of iterations, , then after the first round of iteration, P1 can be obtained.

[0076] At the same time, E t It represents the population P t The corresponding objective function matrix, this step is to merge the objective function matrix of the offspring and the population. It means that it can be replaced. After all population cycles are completed, the target optimization individual that meets the maximum number of cycles can be used as the optimization solution of the target decision function.

[0077] Through the multi-party multi-objective optimization method for drone path planning disclosed in this embodiment, an efficient optimization algorithm capable of solving multiple objectives containing multiple decision parties (objective decision functions) is designed. Among them, providing guidance from the perspective of a target decision function leads to a significant increase in the number of individuals in the population that can provide information. Therefore, in the multi-party multi-objective optimization problem, according to this strategy, most individuals in the population may be able to provide guidance information. Therefore, compared with the traditional multi-objective optimization algorithm, it has the characteristics of high algorithm solution performance and rich diversity of solution solutions.

[0078] Figure 3 This is a flow chart of another multi-party multi-objective optimization method for UAV path planning disclosed in an embodiment of the present application, including steps 301 and 302.

[0079] 301. At the current moment, each individual to be optimized is sorted by non-dominated order through the efficiency decision function to obtain the efficiency domination level; at the current moment, each individual to be optimized is sorted by non-dominated order through the safety decision function to obtain the safety domination level.

[0080] In this embodiment, the multi-party non-dominated sorting operator MPNDS2 is mainly used. The sorting operation takes the number of non-dominated layers under different decision makers DMs as the target, and two rounds of non-dominated ranking are used to calculate the non-dominated layers of the multi-objective optimization problem MPMOPs. Specifically, in this algorithm, the current population is first non-dominatedly sorted according to the efficiency objective function and the safety objective function to obtain the efficiency dominance level and the safety dominance level.

[0081] 302. According to the efficiency dominance level and the safety dominance level, each individual to be optimized at the current moment is sorted in a non-dominated manner to obtain the non-dominated level of all individuals to be optimized at the next moment.

[0082] Therefore, based on step 301, the two sorting results of efficiency dominance level and safety dominance level can be combined into a multi-objective problem, and the individuals to be optimized in the initial population are again non-dominated sorted to obtain the non-dominated level of the individuals to be optimized, that is, the population sorting result at the next moment.

[0083] In this embodiment, in combination with step 301 and step 302, MPNDS2 can be expressed in pseudo code form:

[0084]

[0085] Combined with the above description, we can see that in this algorithm, first, P t (Current initial population) according to (efficiency objective function) and (Safety objective function) is sorted by non-dominated sorting, and we get (Efficiency Dominance Level) and Then, these two ranking results are combined into a multi-objective problem and then Perform non-dominated sorting and get , that is, the population sorting result at the next moment.

[0086] It should be noted in advance that the activation operation is used in multi-party multi-objective immune algorithms MOIAs. The sorting results of MPNDS2 are used to replace the results of fast non-dominated sorting. Unlike these ordinary multi-objective immune optimization algorithms that only activate the first non-dominated layer, in multi-party multi-objective problems MPMOPs, the first non-dominated layer individuals sorted by MPNDS2 usually only account for a small part of the population. Therefore, all individuals with a sorting level less than nA are activated.

[0087] .

[0088] in, The population P is The sorted individual set, nA is the preset number of activated individuals. For details, please refer to the embodiment shown in 4.

[0089] In the multi-party multi-objective optimization method for drone path planning disclosed in this embodiment, the sorting operation regards the number of non-dominated layers under different decision makers (DMs) as the target, and two rounds of non-dominated ranking are used to calculate the non-dominated layers for solving multi-objective optimization problems (MPMOPs). By combining the sorting results of these different objective functions into a multi-objective problem, and performing non-dominated sorting on the population again, that is, the population sorting result at the next moment, the preferences of different decision makers are taken into account, and the effectiveness and robustness of the algorithm are ensured through two rounds of sorting.

[0090] Figure 4 This is a flow chart of another multi-party multi-objective optimization method for UAV path planning disclosed in an embodiment of the present application, including steps 401 to 404.

[0091] 401. Define the activation number value of the optimized individuals to be activated, and set the activation number value to the number of activated optimized individuals preset in the activated optimized individual set, so as to obtain the set of optimized individuals to be activated according to the preset number of activated optimized individuals and the individuals to be optimized that are ranked high in the initial population.

[0092] It should be noted that the multi-party adaptive activation operator algorithm AdaptiveActivation is used in this embodiment. Among them, the activation operation in the algorithm is to select a certain proportion of high-quality individuals from the population for cloning and generate improved individuals. Therefore, in order to achieve an overly large set that makes it difficult for the algorithm to effectively distinguish more convergent individuals, it is necessary to define the activation number value of the optimized individuals to be activated, and set the activation number value to the number of activated optimized individuals preset in the activated optimized individual set, so as to obtain the set of optimized individuals to be activated according to the preset number of activated optimized individuals and the top-ranked individuals to be optimized in the initial population.

[0093] In one specific embodiment, the minimum activation amount and the activation amount value mentioned above are first determined from the candidate activation amount list, and then the list element corresponding to the activation amount value is set to the size of the optimized individual to be activated in the activation optimization individual set. Thus, the optimized individual set to be activated can be obtained.

[0094] 402. Calculate the Pareto front individual set on the Pareto front for any objective decision function.

[0095] Corresponding to step 401, it is also necessary to calculate the Pareto front individual set on the Pareto front of any target decision function. It should be noted that the Pareto front individual set includes the individuals to be optimized with the highest ranking.

[0096] It should be noted that the boundary individuals of the Pareto front are usually part of the Pareto non-dominated set. In one specific embodiment, the set of individuals located on the Pareto front of any target decision function (decision maker) among all individuals to be activated is calculated, thereby obtaining the Pareto front individual set. Further, for example, under certain conditions, in the DM1 perspective, its Pareto front individual set is From the perspective of DM2, the Pareto frontier individuals are .

[0097] 403. Calculate the target value component of each individual to be optimized in the set of optimized individuals to be activated and the Pareto frontier individual set, and obtain coverage metrics corresponding to different target decision functions.

[0098] Based on step 401 and step 402, the target value component of each individual to be optimized in the set of individuals to be activated and optimized and the Pareto frontier individual set can be calculated, thereby obtaining coverage metrics corresponding to different target decision functions.

[0099] In one of the specific embodiments, in order to solve the problem that a fixed-size activation strategy may lead to a loss of diversity when dealing with multi-party multi-objective optimization problems MPMOPs, a value called multi-party coverage metric (MCM) is used to evaluate the performance of different sets of activated individuals. Before calculating MCM, it is necessary to calculate the coverage metric (cm) for different decision makers.

[0100] Specifically, the coverage metric (cm k ) is calculated using the following formula:

[0101] It should be noted that the function represents the target value component of the i-th individual to be optimized in the corresponding individual set. A represents the given activated individual set, and B k M is a set of individuals located on the Pareto front of the kth decision maker among all individuals. k represents the target set of the kth decision maker. In short, the set B k represents the set of individuals on the Pareto front (PF) of the kth DM, and M k represents the target set of the kth DM.

[0102] Coverage measurement cm k The calculation of involves the average of the minimum distances between each individual in the population and the nearest individual on the PF of the kth DM. In this example, it is necessary to calculate for each individual x in A i With B k The minimum distance between the closest individuals in A and the average of these distances. For the first DM, since the individuals in A If all individuals in A are in B1, cm1 will be small because the individuals in A are very close to those in B1, which indicates that A lacks diversity from the perspective of the first DM. For the second DM, all individuals in A are in B2, so cm2 will be close to 1, which indicates that A has sufficient diversity from the perspective of the second DM. This example intuitively shows how the coverage metric cm can measure the coverage of the set of activated individuals A from the perspective of different DMs. cm2 is close to 1 while cm1 is much smaller than cm2, which indicates that the given set of activated individuals A has sufficient diversity for DM2 but lacks diversity for DM1.

[0103] 404. If the coverage metrics of different target decision functions are greater than a preset metric threshold, all target coverage metrics greater than the preset metric threshold are obtained, and the top-ranked individuals to be optimized corresponding to the target coverage metrics are determined to obtain an activated optimization individual set.

[0104] Therefore, based on step 403, since a preset metric threshold is set to balance diversity and convergence, when the coverage metrics of different target decision functions are greater than the preset metric threshold, all target coverage metrics greater than the preset metric threshold can be obtained, and the top-ranked individuals to be optimized corresponding to the target coverage metrics can be determined, thereby obtaining an activated optimized individual set.

[0105] In one specific embodiment, after calculating the coverage metric cm of all decision makers, the smallest cm is selected as an indicator to measure the quality of a given set of activated individuals. This indicator is called multi-party coverage metric (MCM), and its calculation formula is as follows:

[0106] , where cm k is the coverage measure of the kth decision maker, and K is the number of decision makers. MCM measures the proportion of the Pareto front distribution range of a given set of activated individuals A from the perspective of the worst-case decision maker. The higher the MCM value, the better the diversity of the given set of activated individuals A among all decision makers. To further illustrate the logic of the multi-party adaptive activation algorithm AdaptiveActivation, the following pseudo code is given:

[0107]

[0108] Combined with the above description, it can be seen that by calculating the MCM of a given set of activated individuals A, the diversity of A from the perspective of different decision makers can be quantified. The goal of the algorithm is to select an activated individual set A with a high MCM value to maintain sufficient diversity in the offspring. As the size of the activated individual set A increases, the MCM always increases monotonically, but a set that is too large may make it difficult for the algorithm to effectively distinguish more convergent individuals, thereby affecting the convergence performance of the algorithm for MPMOPs. Therefore, a threshold (such as 0.99) is set to balance diversity and convergence. Select the minimum number of activations that ensures that the MCM is not less than the threshold from the candidate activation number list ActivateNumList (for example, {20,30,40,50,60,70}).

[0109] Through the multi-party multi-objective optimization method for drone path planning disclosed in this embodiment, the cloning operator is a key component of the immune algorithm, which allocates more replication resources to activated individuals with higher quality, increasing their chances of cross-mutation and producing offspring. In many immune algorithms, resource allocation is mainly determined by the crowding distance. This algorithm fully considers that some activated individuals may not be the Pareto solution of MPMOP, thereby effectively allocating computing resources by utilizing the crowding distance and combining the convergence metric while maintaining population convergence.

[0110] Figure 5 This is a flow chart of another multi-party multi-objective optimization method for UAV path planning disclosed in an embodiment of the present application, including steps 501 to 507.

[0111] 501. Determine the individual quantity value of the optimized individuals to be cloned, and obtain a convergence index corresponding to each activated optimized individual according to the non-dominated sorting of each activated optimized individual in the non-dominated level in the activated optimized individual set.

[0112] It should be noted that the present embodiment uses a cloning operator algorithm and an adaptive operator algorithm. Specifically, it is necessary to first determine the number of optimized individuals to be cloned, and then obtain a convergence index corresponding to each activated optimized individual according to the non-dominated sorting of each activated optimized individual in the non-dominated level in the activated optimized individual set. Specifically, steps 501 to 503 use a cloning operator algorithm, and steps 504 to 507 use an adaptive operator algorithm.

[0113] In one of the specific embodiments, the allocation of cloned individuals is determined by multiple non-dominated sorting layers and crowding distance information. However, in multi-layer non-dominated sorting, individuals with larger convergence indicators have better convergence. Therefore, it is necessary to define the number of optimized individuals to be cloned. It should be understood that at this time, the number of optimized individuals to be cloned can be understood as the preset number of activated individuals, because the optimized individuals to be cloned at this time are derived from the activated optimized individuals. Therefore, after defining the individual number value of the optimized individuals to be cloned, the indicator convergence can be further achieved. Furthermore, the convergence measure p of the i-th activated individual i It can be calculated using the following formula:

[0114] , where p i represents the convergence measure of the i-th activated individual, i.e., the convergence index, nA is the number of activated individuals, layer i represents the multi-party non-dominated ranking layer of the i-th activated individual. i The higher the value of , the better the convergence.

[0115] Furthermore, in the extended multi-party multi-objective immune algorithm, in order to increase the convergence pressure of the algorithm when solving the multi-party multi-objective optimization problem, the number of multi-party non-dominated layers is also used as one of the indicators of resource allocation to calculate the number of clones. The formula of another convergence indicator proposed in this embodiment is as follows:

[0116] , where AF i represents the multi-layer non-dominated sorting of the i-th activated individual. Individuals with larger convergence index have better convergence.

[0117] 502. Obtain a clone quantity value corresponding to each optimized individual to be cloned according to the crowding distance information, the convergence index and the individual quantity value of each activated optimized individual.

[0118] Based on step 501, the crowding distance information, convergence index and individual quantity value of each activated optimized individual can be calculated to obtain the clone quantity value corresponding to each optimized individual to be cloned. It should be noted that the optimized individual to be cloned is one or more of the activated optimized individuals in the activated optimized individual set.

[0119] In one specific embodiment, based on p i , determine the number of clones for each individual. Among them, there is a formula as follows: ,

[0120] in, represents the number of copies of the i-th activated individual, A i represents the i-th activated individual, represents the corresponding crowding distance, It is a rounded-up function, and nC represents the total number of individuals that one wishes to clone.

[0121] Combined with the formula of another convergence indicator in step 501 above, another formula for calculating the number of clones of each individual is as follows: ,

[0122] Among them, Nums i represents the number of copies of the i-th activated individual, a i represents the i-th activated individual, function represents the corresponding crowding distance, is the crowding distance of the i-th activated individual.

[0123] 503. According to the optimized individuals to be cloned and the clone quantity value, a set of individuals to be cloned is obtained, and the set of individuals to be cloned is merged with the activated optimized individual set to obtain a cloned optimized individual set.

[0124] Therefore, based on step 502, the set of individuals to be cloned can be obtained from the optimized individuals to be cloned and the clone quantity value, so that the set of individuals to be cloned and the activated optimized individual set are merged to obtain the cloned optimized individual set.

[0125] In one specific embodiment, after determining the clone number value, the correspondence of the optimized individuals to be cloned and the activated optimized individuals are screened based on the clone number value, and then the two sets after screening are merged to obtain a clone optimized individual set.

[0126] To further illustrate the logic of the cloning operator algorithm Clone, the following pseudo code is provided:

[0127]

[0128] It should be noted that this method ensures that the MPIA algorithm considers diversity and sufficient convergence during the cloning process. When all activated individuals are in the first multi-party non-dominated sorting layer of the MPMOP, this method is equivalent to the ordinary cloning method. At the same time, it can be seen from the third to fifth columns of the above algorithm that it is the pseudo code form of step 503 in this embodiment, and the specific details are not repeated here.

[0129] 504. Determine the maximum number of iterations corresponding to the multi-party adaptive crossover operator algorithm, and record the current number of iterations corresponding to the initial population based on the current number of evaluations, so as to obtain the operator probability of the multi-party adaptive crossover operator algorithm according to the maximum number of iterations and the current number of iterations, so as to select any one of the large step size algorithm, the simulated binary crossover algorithm or the differential operator algorithm according to the operator probability of the multi-party adaptive crossover operator algorithm.

[0130] It should be noted that the multi-party adaptive crossover operator algorithm includes at least three operators. Operator 1: rand / 2 / binDE operator, which mainly uses a large step size to enable the algorithm to show strong search capabilities in the early stages. Operator 2: Use simulated binary crossover to perform crossover bootstrap operations. Operator 3: The commonly used rand / 1 / bin DE operator (differential operator algorithm), which is used when the individual does not meet the conditions of operator 1 and operator 2. This is a basic DE operator.

[0131] In this embodiment, it is necessary to determine the maximum number of iterations of the current multi-party adaptive crossover operator algorithm, so as to record the current number of iterations corresponding to the initial population based on the current number of evaluations. Then, according to the maximum number of iterations and the current number of iterations, the operator probability of the multi-party adaptive crossover operator algorithm is obtained, so as to select any one of the large step size algorithm, the simulated binary crossover algorithm or the differential operator algorithm according to the operator probability of the multi-party adaptive crossover operator algorithm.

[0132] In one specific embodiment, the operator probabilities of the multi-party adaptive crossover operator algorithm are first defined for cross compilation. Specifically, the first operator is defined with probability Used for crossover mutation. The probability of using the first operator is controlled by the following formula, which is related to the number of iterations: , where t represents the current iteration number (current generation) and T represents the maximum iteration number (maximum generation). This control method enables the algorithm to search with a large step size in the early stage, accelerates the convergence of the population and enhances the algorithm's search ability. Then based on different and conditions, and execute any one of step 505, step 506 or step 507 respectively.

[0133] Furthermore, in terms of crossover, the simulated binary crossover (SBX) operator can be used, or the SBX operator or the differential evolution operator (DE2) can be used with a certain probability. In addition, the adaptive multiple differential evolution operators (DE1, DE2 and DE3) can also be used, where DE1, DE2 or DE3. In terms of mutation, all algorithms use the polynomial mutation operator (PM).

[0134] 505. If the operator probability of the multi-party adaptive crossover operator algorithm is greater than the random probability value randomly generated in the multi-party adaptive crossover operator algorithm, determine the individual sizes of all activated optimized individuals in the cloned optimized individual set, and determine to use a large step size to make the algorithm.

[0135] In this embodiment, when the operator probability of the multi-party adaptive crossover operator algorithm is greater than the random probability value randomly generated in the multi-party adaptive crossover operator algorithm, step 505 is executed. Specifically, the individual sizes of all activated optimized individuals in the clone optimized individual set are first determined; then the large step size is defined to make the first scaling factor, the first crossover rate and the decision variable corresponding to each clone optimized individual in the clone optimized individual set in the algorithm, and a predetermined number of activated optimized individuals are arbitrarily selected from the individual sizes of all activated optimized individuals, so that the individual sizes corresponding to the predetermined number of activated optimized individuals are used as index values ​​of the activated optimized individuals; wherein the index value includes at least the individual sizes corresponding to four groups of activated optimized individuals. Then, when the operator probability of the multi-party adaptive crossover operator algorithm is less than the first crossover rate, any clone optimization individual is determined in the clone optimization individual set, and the offspring optimization individual corresponding to any clone optimization individual, the first scaling factor and the activated optimization individual corresponding to any two groups of index values ​​is obtained; wherein the first crossover rate is less than the preset probability threshold; or, when the operator probability of the multi-party adaptive crossover operator algorithm is equal to the decision variable corresponding to any clone optimization individual, the offspring optimization individual corresponding to any clone optimization individual is obtained according to the first scaling factor and the activated optimization individual corresponding to the other two groups of index values; wherein the activated optimization individuals corresponding to the other two groups of index values ​​are other activated optimization individuals except the activated optimization individuals corresponding to any two groups of index values; or, when the operator probability of the multi-party adaptive crossover operator algorithm is greater than the first crossover rate and is not equal to the decision variable corresponding to any clone optimization individual, any clone optimization individual is determined to be the offspring optimization individual.

[0136] In one specific embodiment, when the operator probability of the multi-party adaptive crossover operator algorithm determined in step 504 is greater than the random probability value rand randomly generated in the multi-party adaptive crossover operator algorithm, the operator op1, i.e., the rand / 2 / bin operator, is executed. Specifically, in combination with the actual steps described above, the calculation is performed by the following formula: , where O represents the offspring individual after the operation, C represents the cloned individual, and A represents the activated individual. The subscripts i and j represent the jth decision variable of the i-th corresponding individual. The indices r1, r2, r3, and r4 are from A unique integer randomly selected from nA, nA is the size of the activation individual. CR1 represents the crossover rate (first crossover rate), which is 0.9. F1 represents the scaling factor (first scaling factor), which is 0.7.

[0137] Furthermore, in other feasible technical solutions, the algorithm is based on probability Use operator 2 and operator 3. Three conditions when using operator 2 (cross-guide strategy operator) If the condition is met, the algorithm uses operator 2 op2. Otherwise, the algorithm switches to operator 3 op3. It should be noted that the three conditions of operator 2 (cross-guidance strategy operator) are As follows: Con1: The cloned individuals (guided individuals) are not located at the Pareto non-dominated frontier of multiple parties after sorting by the MPNDS2 operator. This ensures that the generated offspring tend to be at the frontier of multiple decision makers.

[0138] Con2: There are individuals in the population that are closer or as close to the frontier as the guided individuals, and the multi-party Pareto non-dominated ranking of the guiding individuals is smaller than that of the guided individuals. This enables the guiding individuals to provide effective information to the guided individuals.

[0139] Con3: The guided individual and the corresponding clone have never undergone a guided crossover, then the probability Use guided crossover. This allows individuals and their corresponding clones to use a variety of different crossover methods to maintain population diversity.

[0140] For the convenience of description, this will not be repeated later.

[0141] 506. If the operator probability of the multi-party adaptive crossover operator algorithm is less than the random probability value, and the cloned optimized individual meets the first screening condition and the second screening condition, it is determined to use the simulated binary crossover algorithm.

[0142] In this embodiment, when the operator probability of the multi-party adaptive crossover operator algorithm is less than the random probability value, and the cloned optimized individual meets the first screening condition and the second screening condition, the second crossover rate and the guide index value of the activated optimized individual corresponding to any cloned optimized individual are defined based on the simulated binary crossover algorithm; wherein the second crossover rate is the inverse of the decision variable corresponding to each cloned optimized individual in the cloned optimized individual set, and the second crossover rate is greater than a preset probability threshold; when the operator probability of the multi-party adaptive crossover operator algorithm is less than the second crossover rate, the offspring optimized individual corresponding to any cloned optimized individual and the activated optimized individual corresponding to the guide index value is obtained; or, when the operator probability of the multi-party adaptive crossover operator algorithm is greater than or equal to the second crossover rate, any cloned optimized individual is determined to be a offspring optimized individual.

[0143] In one specific embodiment, a simulated binary crossover may be used to perform the crossover boot operation. Specifically, if and only if the above step 505 When all are satisfied, step 506 is executed. It is necessary to find a guiding individual that satisfies Con1 and Con2 in the activated individual population. If and only if there is a guiding individual that satisfies Con1 and Con2, and Con3 is satisfied at the same time, step 506 is executed and implemented by the following formula: , where O represents the offspring individual after the operation, C represents the cloned individual, and A represents the activated individual. The subscript i,j represents the jth decision variable of the i-th corresponding individual. The subscript guide is the index of the guide individual. c represents the crossover rate, which is the inverse of the decision variable. Calculated based on the crossover distance disM, where disC is the crossover distance.

[0144] 507. If the cloned optimized individual meets the third screening condition, determine to use the differential operator algorithm.

[0145] In this embodiment, when the cloned optimized individual meets the third screening condition, the second scaling factor and the third crossover rate in the differential operator algorithm are defined; thus, when the operator probability of the multi-party adaptive crossover operator algorithm is less than the third crossover rate, any cloned optimized individual is determined in the cloned optimized individual set, and based on any cloned optimized individual, the second scaling factor and the activated optimized individual corresponding to any two sets of index values, the offspring optimized individual corresponding to any cloned optimized individual is obtained; or, when the operator probability of the multi-party adaptive crossover operator algorithm is greater than the third crossover rate, any cloned optimized individual is determined as a offspring optimized individual.

[0146] In one specific embodiment, step 507 mainly uses the commonly used rand / 1 / bin DE operator, which is used when the individual does not meet the conditions of step 505 or step 506. Specifically, if and only if the (Con 1， Con 2， Con3) are not satisfied, step 507 is executed. That is, no guiding individual satisfying Con1 and Con2 is found in the activated individual population, and when Con3 is not satisfied, step 507 is executed and implemented by the following formula: , where CR2 is the crossover rate, which is 0.5. F2 is the scaling factor, which is 0.5.

[0147] Furthermore, after all crossovers, the individuals are mutated using the polynomial mutation (PM) operator.

[0148] In combination with the above steps 504 to 507, under certain conditions:

[0149]

[0150] Among them, O represents the individual generated after the operation, C represents the cloned individual, and A 1 -A 4represents different activated individuals. Subscripts i, j represent the jth decision variable or corresponding offspring of the i-th individual. CR1-CR3 represent crossover rates, which are 0.9, 0.5, and 0.1, respectively. F1-F3 are scaling factors, which are 0.7, 0.5, and 0.1, respectively. P c is the crossover rate, which is the inverse of the decision variable. Calculated based on the crossover distance disM, where disC is the crossover distance.

[0151] Furthermore, in combination with the above steps 504 to 507, in order to illustrate the operation logic of the multi-party adaptive crossover operator AdaptiveOperator, a pseudo code description is given below:

[0152]

[0153] Through the multi-party multi-objective optimization method for drone path planning disclosed in this embodiment, it is ensured that the MPIA algorithm considers diversity and sufficient convergence in the cloning process. Furthermore, when all activated individuals are in the first multi-party non-dominated sorting layer of MPMOP, this method is equivalent to the ordinary cloning method, which improves the applicability of the algorithm. At the same time, a set of rules is designed to ensure the appropriate application of two differential evolution (DE) operators and the cross-guidance strategy proposed in the algorithm during the algorithm process, which improves the feasibility of the scheme.

[0154] Figure 6 This is a flow chart of another multi-party multi-objective optimization method for UAV path planning disclosed in an embodiment of the present application, including steps 601 to 603.

[0155] 601. Determine the target number value of all clone optimization individuals in the candidate population.

[0156] It should be noted that the selection operator algorithm is used in this embodiment. Specifically, as described in step 206, the target number of all clone optimization individuals in the candidate population needs to be determined in advance.

[0157] 602. Based on the selection operator algorithm, sort each clone optimization individual in the candidate population according to the crowding distance information, add the clone optimization individuals in the current non-dominated sorting layer to the target population, and execute the sorting step of the next non-dominated sorting layer.

[0158] Corresponding to step 601, an operator algorithm can be selected, and then each clone optimization individual in the candidate population can be sorted according to the crowding distance information, and the clone optimization individuals in the current non-dominated sorting layer can be added to the target population, and a loop is executed thereby, so that the individuals in the next layer can be added to the target population. It should be noted that, in this embodiment, the current non-dominated sorting layer is a sorting layer generated after non-dominated sorting by any target decision function, and any non-dominated sorting layer includes the candidate population obtained after using any target decision function.

[0159] In one specific embodiment, each layer of individuals in the candidate is sorted according to the crowding distance (the sorting can be performed using the MPNDS2 operator algorithm), so that the cloned preferred individuals at the top of the sort (low rank) are screened out, and the screened cloned preferred individuals are determined as the target population.

[0160] Furthermore, when the number of clone optimization individuals retained in the target population is less than the target number, the clone optimization individuals in the current non-dominated sorting layer are added to the target population, until the determination condition of step 603 is met.

[0161] 603. When the number of cloned optimized individuals retained in the target population is greater than the target number, the cloned optimized individuals with the maximum crowding distance information are removed from the target population according to the crowding distance information, and the remaining cloned optimized individuals in the target population are determined as target optimized individuals.

[0162] Therefore, based on step 602, when the number of cloned optimized individuals retained in the target population is greater than the target number, it is necessary to remove the cloned optimized individuals with the maximum crowding distance information in the target population according to the crowding distance information, and at the same time, determine the remaining cloned optimized individuals in the target population as target optimized individuals.

[0163] In one specific embodiment, when adding the cloned optimized individuals to the target population, the number of cloned optimized individuals in the current target population is determined. When the number of cloned optimized individuals begins to be greater than the predetermined number of cloned individuals, the most crowded cloned optimized individuals in the current non-dominated sorting layer are removed from the target population until the screening of cloned optimized individuals in all layers is completed.

[0164] To further illustrate the logic of the selection operator algorithm Selection, the following pseudo code is provided:

[0165]

[0166] Through the multi-party multi-objective optimization method for drone path planning disclosed in this embodiment, solutions are retained layer by layer according to the number of layers of the multi-party non-dominated sorting operator MPNDS2. In a given layer, if the total number of retained solutions exceeds the number of maintained populations, the crowding distance information will filter out some crowded solutions in the layer, which is conducive to maintaining the diversity of the population.

[0167] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0168] See also Fig.10 , Fig.10 A schematic diagram of the structure of a multi-party multi-objective optimization device disclosed in an embodiment of the present application.

[0169] The acquisition unit 1001 is used to acquire the initial population in the objective function space corresponding to the planned path of the drone of the multi-objective decision function, and perform non-dominated sorting on each individual to be optimized in the initial population to obtain the non-dominated level of all individuals to be optimized; wherein the multi-objective decision function includes an efficiency decision function and a safety decision function, the objective function space includes a function matrix of the planned path of the drone from the perspective of the efficiency decision function, and a function matrix of the planned path of the drone from the perspective of the safety decision function, and the initial population includes an initial function solution generated by the efficiency decision function and the safety decision function in the objective function space;

[0170] The sorting unit 1002 is used for sorting according to the non-dominated level when the current evaluation times of any objective decision function do not meet the preset evaluation times, and activating the individuals to be optimized with the top non-dominated ranking according to the multi-party adaptive activation operator algorithm to obtain an activated optimized individual set; wherein the activated optimized individual set includes the activated optimized individual, and the activated optimized individual is used to characterize the individuals to be optimized after the adaptive activation;

[0171] The calculation unit 1003 is used to calculate the crowding distance information of each activated optimization individual in the activated optimization individual set according to the clonal operator algorithm, so as to allocate resources in the activated optimization individual set according to the crowding distance information and the non-dominated level corresponding to each activated optimization individual, and obtain the cloned optimization individual set; wherein the cloned optimization individual set includes the cloned optimization individuals, and the cloned optimization individuals are used to represent the activated optimization individuals after resource allocation;

[0172] A generating unit 1004 is used to generate offspring optimized individuals according to all clone optimized individuals in the clone optimized individual set according to a multi-party adaptive crossover operator algorithm;

[0173] A merging unit 1005 is used to merge the offspring optimized individuals with the initial population to obtain a candidate population;

[0174] The selection unit 1006 is used to sort the candidate population according to the selection operator algorithm and the crowding distance information, and select the target optimized individual from the sorted candidate population as the population participating in the next round of cycles until the maximum number of cycles is reached, and the target optimized individual that meets the maximum number of cycles is used as the optimization solution of the target decision function; wherein the target optimized individual that reaches the maximum number of cycles is the target function solution of the efficiency decision function and the safety decision function in the target function space corresponding to the planned path of the drone.

[0175] See below Fig.11 , a schematic diagram of the structure of a multi-party multi-objective optimization device disclosed in an embodiment of the present application includes:

[0176] CPU 1101, memory 1105, input / output interface 1104, wired or wireless network interface 1103 and power supply 1102;

[0177] The memory 1105 is a temporary storage memory or a permanent storage memory;

[0178] The CPU 1101 is configured to communicate with the memory 1105 and execute the instructions in the memory 1105 to perform the aforementioned Figures 2 to 6 The multi-party and multi-objective optimization method for UAV path planning in the illustrated embodiment.

[0179] The embodiment of the present application also provides a chip system, the chip system includes at least one processor and a communication interface, the communication interface and the at least one processor are interconnected through a line, and the at least one processor is used to run a computer program or instruction to execute the aforementioned Figures 2 to 6 The multi-party and multi-objective optimization method for UAV path planning in the illustrated embodiment.

[0180] The embodiment of the present application also provides a computer-readable storage medium, the computer-readable storage medium includes instructions, when the instructions are executed on a computer, the computer executes the aforementioned Figures 2 to 6 The multi-party and multi-objective optimization method for UAV path planning in the illustrated embodiment.

[0181] The present application also provides a computer program product including instructions, which, when executed on a computer, enables the computer to execute the aforementioned Figures 2 to 6 The multi-party and multi-objective optimization method for UAV path planning in the illustrated embodiment.

[0182] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0183] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0184] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0185] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0186] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, read-only memory), random access memory (RAM, random access memory), disk or optical disk and other media that can store program code.

Claims

1. A multi-party multi-objective optimization method for UAV path planning, characterized in that: The method comprises: Obtain an initial population in the objective function space of the multi-objective decision function corresponding to the planned path of the UAV, and perform non-dominated sorting on each individual to be optimized in the initial population to obtain the non-dominated level of all individuals to be optimized; wherein the multi-objective decision function includes an efficiency decision function and a safety decision function, the objective function space includes a function matrix of the planned path of the UAV from the perspective of the efficiency decision function, and a function matrix of the planned path of the UAV from the perspective of the safety decision function, and the initial population includes an initial function solution generated by the efficiency decision function and the safety decision function in the objective function space; the efficiency decision function is associated with one or more of the flight efficiency, flight trajectory length, flight altitude change, flight energy consumption or the sum of mission hovering point distances of the UAV, and the safety decision function is associated with one or more of the fatal risks of pedestrians and moving vehicles of the UAV, the property risks of buildings or the noise pollution risks in urban areas; When the current evaluation times of any objective decision function do not meet the preset evaluation times, the non-dominated levels are sorted, and the individuals to be optimized with the top non-dominated rankings are activated according to the multi-party adaptive activation operator algorithm to obtain an activated optimized individual set; wherein the activated optimized individual set includes the activated optimized individual, and the activated optimized individual is used to characterize the individuals to be optimized after the adaptive activation; According to the clonal operator algorithm, the crowding distance information of each activated optimized individual in the activated optimized individual set is calculated, so that in the activated optimized individual set, resources are allocated according to the crowding distance information and the non-dominated level corresponding to each activated optimized individual to obtain a clonal optimized individual set; wherein the clonal optimized individual set includes clonal optimized individuals, and the clonal optimized individuals are used to characterize the activated optimized individuals after resource allocation; According to a multi-party adaptive crossover operator algorithm, based on all clone optimized individuals in the clone optimized individual set, offspring optimized individuals are generated; Merging the offspring optimized individuals with the initial population to obtain a candidate population; The candidate population is sorted according to the selection operator algorithm and the crowding distance information, and the target optimized individual is selected from the sorted candidate population as the population participating in the next round of cycles until the maximum number of cycles is reached, and the target optimized individual that meets the maximum number of cycles is used as the optimization solution of the target decision function; wherein the target optimized individual that reaches the maximum number of cycles is the target function solution of the efficiency decision function and the safety decision function corresponding to the planned path of the drone in the target function space; The above method activates the non-dominated individuals to be optimized that are ranked higher in the ranking according to the multi-party adaptive activation operator algorithm to obtain an activated optimized individual set, including: Defining an activation quantity value of the optimized individuals to be activated, and setting the activation quantity value to the quantity of the activated optimized individuals preset in the activated optimized individuals set, so as to obtain the set of optimized individuals to be activated according to the preset quantity of the activated optimized individuals and the individuals to be optimized with the highest ranking in the initial population; Calculate the Pareto front individual set on the Pareto front of any objective decision function; wherein the Pareto front individual set includes the individuals to be optimized with the highest ranking; Calculate the target value component of each individual to be optimized in the set of optimized individuals to be activated and the Pareto frontier individual set to obtain coverage metrics corresponding to different target decision functions; If the coverage metrics of different target decision functions are greater than a preset metric threshold, all target coverage metrics greater than the preset metric threshold are obtained, and the top-ranked individuals to be optimized corresponding to the target coverage metrics are determined to obtain the activated optimized individual set; The method of allocating resources in the activated optimization individual set according to the crowding distance information and the non-dominated level corresponding to each activated optimization individual to obtain a clone optimization individual set includes: Determine the individual quantity value of the optimized individuals to be cloned, and obtain a convergence index corresponding to each activated optimized individual according to the non-dominated order of each activated optimized individual in the activated optimized individual set in the non-dominated level; Obtaining a clone quantity value corresponding to each optimized individual to be cloned according to the crowding distance information, the convergence index and the individual quantity value of each activated optimized individual; wherein the optimized individual to be cloned is one or more of the activated optimized individuals in the activated optimized individual set; According to the optimized individuals to be cloned and the clone quantity value, a set of individuals to be cloned is obtained, and the set of individuals to be cloned is merged with the activated optimized individual set to obtain the cloned optimized individual set; The multi-party adaptive crossover operator algorithm includes one or more of a large step size algorithm, a simulated binary crossover algorithm or a differential operator algorithm. According to the multi-party adaptive crossover operator algorithm, based on all the clone optimized individuals in the clone optimized individual set and all the individuals to be optimized in the initial population, before generating offspring optimized individuals, the method includes: Determining a maximum number of iterations corresponding to the multi-party adaptive crossover operator algorithm; Recording a current number of iterations corresponding to the initial population based on the current number of evaluations; According to the maximum number of iterations and the current number of iterations, an operator probability of a multi-party adaptive crossover operator algorithm is obtained, so as to select any one of the large step size algorithm, the simulated binary crossover algorithm or the differential operator algorithm according to the operator probability of the multi-party adaptive crossover operator algorithm; The method of generating offspring optimized individuals according to all clone optimized individuals in the clone optimized individual set according to the multi-party adaptive crossover operator algorithm includes: If the operator probability of the multi-party adaptive crossover operator algorithm is greater than the random probability value randomly generated in the multi-party adaptive crossover operator algorithm, determining the individual sizes of all activated optimized individuals in the cloned optimized individual set; The first scaling factor, the first crossover rate and the decision variable corresponding to each clone optimization individual in the clone optimization individual set are defined in the large step size algorithm, and a predetermined number of activation optimization individuals are arbitrarily selected from the individual sizes of all activation optimization individuals, so as to use the individual sizes corresponding to the predetermined number of activation optimization individuals as index values ​​of the activation optimization individuals; wherein the index value includes at least the individual sizes corresponding to four groups of activation optimization individuals; If the operator probability of the multi-party adaptive crossover operator algorithm is less than the first crossover rate, any clone optimization individual is determined in the clone optimization individual set, and according to the any clone optimization individual, the first scaling factor and the activated optimization individual corresponding to any two groups of index values, a descendant optimization individual corresponding to the any clone optimization individual is obtained; wherein the first crossover rate is less than a preset probability threshold; If the operator probability of the multi-party adaptive crossover operator algorithm is equal to the decision variable corresponding to any one of the clone optimization individuals, the offspring optimization individuals corresponding to any one of the clone optimization individuals are obtained according to the first scaling factor and the activation optimization individuals corresponding to the other two groups of index values; wherein the activation optimization individuals corresponding to the other two groups of index values ​​are other activation optimization individuals except the activation optimization individuals corresponding to the any two groups of index values; If the operator probability of the multi-party adaptive crossover operator algorithm is greater than the first crossover rate and is not equal to the decision variable corresponding to any of the cloned optimized individuals, the any of the cloned optimized individuals is determined to be the offspring optimized individual.

2. The multi-party multi-objective optimization method for UAV path planning according to claim 1 is characterized in that: If the multi-objective decision function includes an efficiency decision function and a safety decision function, performing non-dominated sorting on each individual to be optimized in the initial population to obtain the non-dominated level of all individuals to be optimized includes: At the current moment, each individual to be optimized is non-dominated and sorted by the efficiency decision function to obtain an efficiency domination level; at the current moment, each individual to be optimized is non-dominated and sorted by the safety decision function to obtain a safety domination level; According to the efficiency dominance level and the safety dominance level, each of the individuals to be optimized at the current moment is non-dominated and sorted to obtain the non-dominated level of all the individuals to be optimized at the next moment.

3. The multi-party multi-objective optimization method for UAV path planning according to claim 1 is characterized in that: The method of generating offspring optimized individuals according to all the clone optimized individuals in the clone optimized individual set and all the individuals to be optimized in the initial population according to the multi-party adaptive crossover operator algorithm includes: If the operator probability of the multi-party adaptive crossover operator algorithm is less than the random probability value, and the clone optimization individual satisfies the first screening condition and the second screening condition, a second crossover rate and a guide index value of the activated optimization individual corresponding to any clone optimization individual are defined based on the simulated binary crossover algorithm; wherein the second crossover rate is the inverse of the decision variable corresponding to each clone optimization individual in the clone optimization individual set, and the second crossover rate is greater than the preset probability threshold; If the operator probability of the multi-party adaptive crossover operator algorithm is less than the second crossover rate, obtaining the offspring optimized individual corresponding to any clone optimized individual according to the any clone optimized individual and the activated optimized individual corresponding to the guide index value; If the operator probability of the multi-party adaptive crossover operator algorithm is greater than or equal to the second crossover rate, any one of the cloned optimized individuals is determined as the offspring optimized individual.

4. The multi-party multi-objective optimization method for UAV path planning according to claim 1 is characterized in that: The method of generating offspring optimized individuals according to all the clone optimized individuals in the clone optimized individual set and all the individuals to be optimized in the initial population according to the multi-party adaptive crossover operator algorithm includes: If the clone optimization individual satisfies the third screening condition, defining a second scaling factor and a third crossover rate in the differential operator algorithm; If the operator probability of the multi-party adaptive crossover operator algorithm is less than the third crossover rate, any clone optimization individual is determined in the clone optimization individual set, and according to the any clone optimization individual, the second scaling factor and the activated optimization individual corresponding to any two groups of index values, a descendant optimization individual corresponding to the any clone optimization individual is obtained; If the operator probability of the multi-party adaptive crossover operator algorithm is greater than the third crossover rate, any one of the clone optimized individuals is determined as the offspring optimized individual.

5. The multi-party multi-objective optimization method for UAV path planning according to claim 1 is characterized in that: The step of sorting the candidate population according to the selection operator algorithm and the crowding distance information, and selecting a target optimization individual from the sorted candidate population, includes: Determining a target number value of all cloned optimized individuals in the candidate population; Based on the selection operator algorithm, each clone optimization individual in the candidate population is sorted according to the crowding distance information, the clone optimization individuals in the current non-dominated sorting layer are added to the target population, and the sorting step of the next non-dominated sorting layer is performed; wherein the current non-dominated sorting layer is a sorting layer generated after non-dominated sorting by any target decision function, and any non-dominated sorting layer includes the candidate population obtained by using any target decision function; When the number of cloned optimized individuals retained in the target population is greater than the target number, the cloned optimized individuals with the maximum crowding distance information are removed from the target population according to the crowding distance information, and the remaining cloned optimized individuals in the target population are determined as the target optimized individuals.

6. A multi-party multi-objective optimization device for UAV path planning, characterized in that: The device comprises: An acquisition unit is used to acquire an initial population in an objective function space corresponding to a multi-objective decision function of a planned path of a UAV, and to perform non-dominated sorting on each individual to be optimized in the initial population to obtain a non-dominated level of all individuals to be optimized; wherein the multi-objective decision function includes an efficiency decision function and a safety decision function, the objective function space includes a function matrix of the planned path of the UAV from the perspective of the efficiency decision function, and a function matrix of the planned path of the UAV from the perspective of the safety decision function, and the initial population includes an initial function solution generated by the efficiency decision function and the safety decision function in the objective function space; the efficiency decision function is associated with one or more of the flight efficiency, flight trajectory length, flight altitude change, flight energy consumption, or total mission hovering point distance of the UAV, and the safety decision function is associated with one or more of the fatal risk of pedestrians and moving vehicles of the UAV, the property risk of buildings, or the noise pollution risk of urban areas; A sorting unit is used for sorting according to the non-dominated level when the current evaluation times of any objective decision function do not meet the preset evaluation times, and activating the individuals to be optimized with the top non-dominated ranking according to the multi-party adaptive activation operator algorithm to obtain an activated optimized individual set; wherein the activated optimized individual set includes the activated optimized individual, and the activated optimized individual is used to characterize the individuals to be optimized after the adaptive activation; A calculation unit, configured to calculate the crowding distance information of each activated optimization individual in the activated optimization individual set according to the clonal operator algorithm, so as to allocate resources in the activated optimization individual set according to the crowding distance information and the non-dominated level corresponding to each activated optimization individual, and obtain a clonal optimization individual set; wherein the clonal optimization individual set includes clonal optimization individuals, and the clonal optimization individuals are used to characterize the activated optimization individuals after resource allocation; A generating unit, configured to generate offspring optimized individuals according to all clone optimized individuals in the clone optimized individual set according to a multi-party adaptive crossover operator algorithm; A merging unit, used for merging the offspring optimized individuals with the initial population to obtain a candidate population; A selection unit is used to sort the candidate population according to the selection operator algorithm and the crowding distance information, and select the target optimized individual from the sorted candidate population as the population participating in the next round of cycles until the maximum number of cycles is reached, and the target optimized individual that meets the maximum number of cycles is used as the optimization solution of the target decision function; wherein the target optimized individual that reaches the maximum number of cycles is the target function solution of the efficiency decision function and the safety decision function corresponding to the planned path of the drone in the target function space; The sorting unit is further used to define an activation quantity value of the optimized individuals to be activated, and set the activation quantity value to the quantity of the activated optimized individuals preset in the activated optimized individuals set, so as to obtain the set of optimized individuals to be activated according to the preset quantity of the activated optimized individuals and the individuals to be optimized with the highest ranking in the initial population; Calculate the Pareto front individual set on the Pareto front of any objective decision function; wherein the Pareto front individual set includes the individuals to be optimized with the highest ranking; Calculate the target value component of each individual to be optimized in the set of optimized individuals to be activated and the Pareto frontier individual set to obtain coverage metrics corresponding to different target decision functions; If the coverage metrics of different target decision functions are greater than a preset metric threshold, all target coverage metrics greater than the preset metric threshold are obtained, and the top-ranked individuals to be optimized corresponding to the target coverage metrics are determined to obtain the activated optimized individual set; The calculation unit is further used to determine the individual quantity value of the optimized individuals to be cloned, and obtain a convergence index corresponding to each activated optimized individual according to the non-dominated ranking of each activated optimized individual in the activated optimized individual set in the non-dominated level; Obtaining a clone quantity value corresponding to each optimized individual to be cloned according to the crowding distance information, the convergence index and the individual quantity value of each activated optimized individual; wherein the optimized individual to be cloned is one or more of the activated optimized individuals in the activated optimized individual set; According to the optimized individuals to be cloned and the clone quantity value, a set of individuals to be cloned is obtained, and the set of individuals to be cloned is merged with the activated optimized individual set to obtain the cloned optimized individual set; The multi-party adaptive crossover operator algorithm includes one or more of a large step size algorithm, a simulated binary crossover algorithm or a differential operator algorithm, and the generating unit is further used to determine a maximum number of iterations corresponding to the multi-party adaptive crossover operator algorithm; Recording a current number of iterations corresponding to the initial population based on the current number of evaluations; According to the maximum number of iterations and the current number of iterations, an operator probability of a multi-party adaptive crossover operator algorithm is obtained, so as to select any one of the large step size algorithm, the simulated binary crossover algorithm or the differential operator algorithm according to the operator probability of the multi-party adaptive crossover operator algorithm; The generating unit is further used to determine the individual sizes of all activated optimized individuals in the clone optimized individual set when the operator probability of the multi-party adaptive crossover operator algorithm is greater than the random probability value randomly generated in the multi-party adaptive crossover operator algorithm; The first scaling factor, the first crossover rate and the decision variable corresponding to each clone optimization individual in the clone optimization individual set are defined in the large step size algorithm, and a predetermined number of activation optimization individuals are arbitrarily selected from the individual sizes of all activation optimization individuals, so as to use the individual sizes corresponding to the predetermined number of activation optimization individuals as index values ​​of the activation optimization individuals; wherein the index value includes at least the individual sizes corresponding to four groups of activation optimization individuals; If the operator probability of the multi-party adaptive crossover operator algorithm is less than the first crossover rate, any clone optimization individual is determined in the clone optimization individual set, and according to the any clone optimization individual, the first scaling factor and the activated optimization individual corresponding to any two groups of index values, a descendant optimization individual corresponding to the any clone optimization individual is obtained; wherein the first crossover rate is less than a preset probability threshold; If the operator probability of the multi-party adaptive crossover operator algorithm is equal to the decision variable corresponding to any one of the clone optimization individuals, the offspring optimization individuals corresponding to any one of the clone optimization individuals are obtained according to the first scaling factor and the activation optimization individuals corresponding to the other two groups of index values; wherein the activation optimization individuals corresponding to the other two groups of index values ​​are other activation optimization individuals except the activation optimization individuals corresponding to the any two groups of index values; If the operator probability of the multi-party adaptive crossover operator algorithm is greater than the first crossover rate and is not equal to the decision variable corresponding to any of the cloned optimized individuals, the any of the cloned optimized individuals is determined to be the offspring optimized individual.

Images