Unmanned aerial vehicle path planning method and system based on chaos multi-agent game optimization
By employing piecewise chaotic mapping, lens imaging inverse learning, and dynamic weight adjustment, combined with sparrow population division of labor and convergence monitoring, the problems of premature convergence and imprecise obstacle modeling in UAV path planning are solved, achieving efficient path planning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-16
- Publication Date
- 2026-03-10
AI Technical Summary
Existing UAV path planning algorithms are prone to premature convergence due to the decrease in initial population randomness and iteration diversity. They also suffer from high missed detection rates in high-dimensional optimal solutions, imprecise obstacle modeling, weak kinematic adaptation, lagging dynamic environment response, and lack of adaptive weight adjustment, all of which affect execution efficiency.
A segmented chaotic mapping is used to generate an initial population. Combined with lens imaging reverse learning and chaotic perturbation monitoring, a three-dimensional constraint model and cylindrical obstacle model are constructed. A dynamic weight adjustment mechanism is designed to simulate the division of labor in a sparrow population. A convergence monitoring unit is integrated to optimize path planning.
It enhances the global exploration capability and local accuracy of UAV path planning, dynamically adapts to complex environments, improves planning response efficiency and execution efficiency, reduces obstacle omissions, and enhances adaptability in complex scenarios.
Smart Images
Figure CN121635408A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the interdisciplinary field of UAV autonomous navigation and intelligent optimization algorithms, specifically involving a UAV path planning method and system based on chaotic multi-agent game optimization. Background Technology
[0002] Unmanned aerial vehicle (UAV) path planning is a core technology for autonomous flight, aiming to find the optimal flight trajectory that meets constraints.
[0003] However, existing technologies have the following shortcomings: traditional intelligent optimization algorithms (such as genetic and particle swarm optimization algorithms) are prone to premature convergence due to the random initial population and the decrease in diversity during iteration; the uneven distribution of the sequence of chaotic solutions in Logistic mapping leads to a high rate of missed detection of high-dimensional optimal solutions; and ant colony and bee colony algorithms are also prone to getting stuck in local short paths in multi-obstacle environments. Most solutions lack fine-grained constraint modeling of the physical characteristics of UAVs, have weak kinematic adaptation, and have crude obstacle avoidance mechanisms. Fixed-weight single-objective optimization, although the NSGA-II algorithm generates Pareto solution sets, lacks adaptive weight adjustment and has a single constraint dimension, which affects execution efficiency; and it has a lag in response in dynamic environments.
[0004] Therefore, it is necessary to develop a path planning system that can effectively combine the advantages of chaotic optimization theory and intelligent algorithms to improve the path planning performance of UAVs in complex environments. Summary of the Invention
[0005] The purpose of this invention is to provide a method and system for UAV path planning based on chaotic multi-agent game optimization. This method enhances global exploration capabilities through segmented chaotic mapping, lens imaging reverse learning, and chaotic disturbance monitoring; it constructs constraint models and hard constraint physical parameters and optimizes obstacle modeling to improve local accuracy; it optimizes multi-objective balance by using dynamic weights and three types of constraints; it simulates sparrow division of labor to speed up the process, adds a monitoring unit to stabilize efficiency, and comprehensively optimizes the performance of UAV path planning.
[0006] To achieve the above objectives, this invention provides a method and system for UAV path planning based on chaotic multi-agent game optimization, comprising the following steps: S1. Construct a three-dimensional comprehensive constraint model that includes path length, flight altitude, and path smoothness, and set hard constraints on the physical parameters of the UAV's maximum turning angle and flight altitude corridor; model obstacles as cylindrical threat zones and introduce penalty terms related to intrusion depth; output the constraint model, physical parameter constraint thresholds, obstacle modeling parameters, and intrusion depth penalty logic. S2. Generate an initial population through segmented chaotic mapping, and generate an inverse solution group by combining lens imaging and inverse learning; monitor the population fitness variance in real time, inject chaotic perturbation when the diversity is below the threshold, and maintain the stability of the chaotic state by monitoring the Lyapunov exponent, and output the initial population, chaotic perturbation trigger threshold and chaotic state monitoring parameters. S3 receives the constraint model from S1 and the initial population from S2, simulates the dynamic role division of finders, followers, and vigilants in a sparrow population, calculates the fitness of individual populations based on the constraint model from S1 and drives the population to iteratively seek optimization, and outputs population dynamic data, convergence state indicators, and the best individual data for each generation. S4 receives the iterative data from S3, analyzes the population distribution and fitness evolution curve in real time through the convergence monitoring unit, dynamically adjusts parameters such as inertia weight, perturbation intensity and crossover probability, and feeds the adjusted parameters back to S2 and S3 to form a closed-loop control. S5 receives the optimal individual data from S3 and the constraint model from S1, performs physical constraints, safety constraints and kinematic feasibility checks on the candidate paths, and outputs the UAV path after the checks are passed.
[0007] Preferably, in S1, the expression for the three-dimensional integrated constraint model is: ; In the formula, As a comprehensive cost; The cost is the path length. The cost of flight altitude; This is the cost of path smoothness; , and Let be the positive weighting coefficients of each cost, and satisfy . .
[0008] Preferably, in S1, the path length cost is the sum of the three-dimensional Euclidean distances of all consecutive path points, and the total journey satisfies... ; In the formula, This is the maximum range of the drone; Total voyage distance; The flight altitude cost is the standard deviation of the waypoint altitude, expressed as: ; ; In the formula, This represents the average flight altitude along the entire route. For the first The flight altitude corresponding to each waypoint; And the height of each path point is full ; In the formula, The minimum permitted flight altitude for drones; This refers to the maximum permitted flight altitude for drones. The path smoothness cost includes a turning angle excess penalty and a climb angle change penalty, expressed as follows: ; In the formula, The weighting coefficient for excessive penalty for turning angle; For the first The climb angle corresponding to each path point; This represents the absolute change in the climb angle between adjacent path segments; The weighting coefficient for the change in the climb angle; This is an excess penalty for turning angles; This represents the total number of path points. The index is the sequence number of the path point; The expression for the intrusion depth penalty term is: ; In the formula, This is the penalty coefficient; The cost of obstacles and threats to the path; Threat radius; The closest distance to the center of the threat.
[0009] Preferably, in S2, the expression for the piecewise chaotic mapping is: ; In the formula, For the first The chaotic value of the next iteration; To control the key parameters of the segmented intervals, .
[0010] Preferably, in S2, the formula for calculating the inverse solution of the lens imaging inverse learning is: ; ; ; ; In the formula, For the first Inverse solution coordinates in each dimension; For the first The first generation of the population The minimum value of the dimension; This is a non-linear scaling factor; For the first The first generation of the population The maximum value of the dimension; For individuals in the first Dimensional coordinates; This represents the total number of iterations. This represents the current iteration number; For the first In the first generation of the population, all individuals in the second generation... A set of coordinates of a dimension; This is the minimum value of the scaling factor; This represents the maximum value of the scaling factor. The formula for the chaotic perturbation is: ; In the formula, For the original position of the individual, The position after the disturbance. For legal operations within the solution space, These are chaotic random numbers generated by piecewise chaotic mapping. The adaptive perturbation strength coefficient is related to the number of iterations. This represents the search scope scale for the current dimension.
[0011] Preferably, in S3, the discoverers account for 20% of the population, their position update strategy includes both safe and insecure scenarios, and the inertial weight expression is: ; ; In the formula, Nonlinear dynamic weights; This is an adaptive step size factor; For the first In the first generation of the population, the first The individual in the first The position coordinates of the dimension; For the first In the first generation of the population, the first The individual in the first The original position coordinates of the dimension; For population alert status parameters; This is a safety threshold; A normally distributed random factor; For the first The optimal individual in the population in the 1st generation The position coordinates of the dimension; These are random numbers distributed according to a standard normal distribution. This is the step size control factor; This represents the minimum value of the inertia weight. This represents the maximum value of the inertia weight; It is a nonlinear decay term; The followers constitute 60%-70% of the population, and their position update strategy is based on random numbers. With cross probability The size relationship determines whether to switch between exploration or mining modes. The expression is: ; In the formula, , For dynamic control parameters; These are uniformly distributed random numbers in the interval 0-1. For exploration factors; For development factors; The crossover probability; For the first The position coordinates of an individual in the population; For the first The original location coordinates of an individual in the population; For the first The coordinates of a high-quality individual (such as the discoverer or the current best individual) in the population; The weighted difference modulus between the superior individual and the current follower's position; The vigilant individuals constitute 10%-20% of the population. Their position update strategy employs different adjustment methods based on the difference between the fitness of an individual and that of a reference individual, as well as the population crowding level. The expression is as follows: ; In the formula, The fitness tolerance threshold, For the disturbance intensity, This is the clustering threshold. For random disturbance factors; This represents the fitness value of an individual. This represents the fitness value of a reference individual in the population. This is the local search step size coefficient; This represents the fitness value of the worst-performing individual in the population. For individual crowding levels.
[0012] Preferably, in S4, the convergence monitoring unit judges the search efficiency by analyzing the population aggregation degree, the growth rate of the optimal fitness and the search trajectory. When a decrease in efficiency is detected, the discoverer inertia weight is increased in the early stage of the iteration to strengthen global exploration, and the follower mining factor is increased in the later stage of the iteration to accelerate local convergence.
[0013] This invention also provides a UAV path planning system based on chaotic multi-agent game optimization, comprising: The constraint model construction module is used to execute S1, build a three-dimensional comprehensive constraint model, complete the modeling of the cylindrical threat zone of the obstacle and set the intrusion depth penalty rule, and output the corresponding evaluation and constraint parameters. The population initialization and diversity maintenance module is connected to the constraint model construction module and is used to execute S2. It generates the initial population and the reverse solution population through piecewise chaotic mapping and lens imaging inverse learning, realizes chaotic disturbance and state monitoring and outputs relevant population data. The multi-agent game iteration module is connected to the population initialization and diversity maintenance module and is used to execute S3 to realize the multi-role division of sparrow population iterative optimization and output iterative data. The parameter dynamic correction module is connected to the multi-agent game iteration module and is used to execute S4 to complete the convergence state analysis and parameter closed-loop control. The path verification output module is connected to the constraint model construction module and the multi-agent game iteration module, respectively, and is used to execute S5 to complete the multi-dimensional verification of the path and output the optimal path.
[0014] Therefore, the present invention employs the above-mentioned UAV path planning method and system based on chaotic multi-agent game optimization. Compared with the prior art, the technical solution of the present invention has the following beneficial effects: (1) The initial population is generated by segmented chaotic mapping, which has better sequence uniformity than Logistic mapping. Inverse learning is combined with lens imaging to generate inverse solution group to ensure chaotic diversity. At the same time, the population fitness variance is monitored in real time. Chaotic perturbation is injected when diversity is insufficient. The Lyapunov exponent is used to monitor and ensure the stability of the chaotic state, thereby strengthening the global exploration capability. (2) Construct a constraint model that includes path length, flight altitude and path smoothness, and impose hard constraints on key physical parameters such as the maximum turning angle and flight altitude corridor of the UAV; in addition, model obstacles as cylindrical threat areas and introduce a penalty term related to intrusion depth to overcome the problem that traditional grid modeling is prone to missing small obstacles, thereby improving the accuracy of local optimization. (3) A dynamic weight adjustment mechanism is designed to adjust and optimize the target in real time according to different task modes without manual intervention; at the same time, it integrates three types of constraints: physical characteristics of UAV, flight environment safety, and core task requirements, which expands the adaptability in complex scenarios and effectively optimizes the multi-objective balance effect. (4) Simulate the dynamic role division of discoverers, followers and vigilants in sparrow population to improve the algorithm's iteration convergence speed and planning response efficiency; it also integrates a convergence monitoring unit to adjust key parameters such as inertia weight and disturbance intensity in real time, so that the optimization efficiency can be maintained without manual debugging when switching scenes.
[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0016] Figure 1 This is an architecture diagram of an embodiment of the UAV path planning method and system based on chaotic multi-agent game optimization of the present invention. Detailed Implementation
[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Unless otherwise defined, the technical or scientific terms used in the present invention should have the ordinary meaning understood by those skilled in the art.
[0018] Example 1 like Figure 1 As shown, this embodiment provides a method and system for UAV path planning based on chaotic multi-agent game optimization. It should be understood that the specific parameters, models and protocols mentioned in this embodiment are only examples to help those skilled in the art understand the present invention, and are not intended to limit the present invention.
[0019] The present invention relates to a UAV path planning method and system based on chaotic multi-agent game optimization, comprising the following steps: S1. Construct a constraint model that includes path length, flight altitude, and path smoothness, and set hard constraints on physical parameters such as the maximum turning angle and flight altitude corridor of the UAV; model obstacles as cylindrical threat zones, introduce penalty terms related to intrusion depth, and output the constraint model, physical parameter constraint thresholds (such as maximum turning angle and altitude corridor range), obstacle modeling parameters, and intrusion depth penalty logic. In this step, a constraint model is constructed using the linear weighted sum method, expressed as: ; In the formula, As a comprehensive cost; The path length cost is directly related to task time and energy consumption. The cost of flight altitude reflects flight stability and the ability to avoid potential threats; The cost of path smoothness is related to the feasibility of drone maneuvering. , and Let be the positive weighting coefficients of each cost, and satisfy . ; The specific configuration of weights can be preset before simulation according to the mission scenario preference, or dynamically assigned by the upper-level mission management system, thereby realizing flexible switching between different mission modes such as reconnaissance, rapid arrival, and covert flight. Assuming the drone's flight path is... It consists of an ordered sequence of path points, denoted as . Each path point Based on its three-dimensional rectangular coordinates definition; The path length cost is calculated by summing the Euclidean distances of all consecutive path points, with the total length constrained to not exceed the UAV's maximum range, and is expressed as: ; In the formula, This represents the total number of path points. The index is the sequence number of the path point; Meanwhile, drones are limited by their fuel capacity or battery capacity, and their total flight path must meet the constraint of the maximum range, expressed as: ; In the formula, For the first The flight path for each segment; This is the maximum range of the drone; Total voyage distance; Paths that do not meet the maximum range constraint will be considered infeasible in the evaluation.
[0020] The flight altitude cost is measured by the standard deviation of the waypoint altitude, which constrains the altitude within the preset flight corridor. On the other hand, to meet the requirements of specific tasks (such as aerial surveying and mapping, and area surveillance) regarding the sensor's field of view and resolution, the UAV needs to maintain a stable flight altitude as much as possible, which can be expressed as: ; ; In the formula, This represents the average flight altitude along the entire route. For the first The flight altitude corresponding to each waypoint; At the same time, each path point height Flight altitude constraints must be met, as shown below: ; In the formula, The minimum permitted flight altitude for drones; This refers to the maximum permitted flight altitude for drones. The path smoothness cost is achieved by penalizing excessive turning angles (yaw angles) and climb / dive angles for any three consecutive points in the path. Its turning angle on the horizontal plane , is represented as: ; ; ; In the formula, For from the first From the path point to the first A planar vector of path points; For from the first From the path point to the first A planar vector of path points; For vectors The modulus length; For vectors The modulus length; At the same time, because drones have a maximum turning angle Due to the constraints, the path smoothness cost is expressed as: ; In the formula, The weighting coefficient for excessive penalty for turning angle; For the first The climb angle corresponding to each path point; This represents the absolute change in the climb angle between adjacent path segments; The weighting coefficient for the change in the climb angle; This is an excess penalty for turning angles; Ground obstacles and no-fly zones are uniformly modeled as cylindrical threat zones, with each threat zone... Define center coordinates Threat radius and buffer distance ; Calculate path segment closest distance to the center of the threat ,like Then in the total cost The intrusion depth penalty term is superimposed and represented as: ; In the formula, This is the penalty coefficient; The cost of obstacles and threats to the path; By optimizing the algorithm to actively avoid threatening areas in the solution space, a safe collision-free path is generated, solving the problem of missing small obstacles in traditional grid modeling; S2. Generate an initial population through segmented chaotic mapping, and generate an inverse solution group by combining lens imaging and inverse learning; monitor the population fitness variance in real time, inject chaotic perturbation when diversity is insufficient, and ensure the stability of the chaotic state by monitoring the Lyapunov exponent, so as to provide high-quality initial population and diversity support for optimization; output the initial population, chaotic perturbation trigger threshold and chaotic state monitoring parameters. In this step, the population quality is improved through chaotic dynamics to avoid initial solution aggregation and premature convergence during iteration. Specifically: Using a piecewise chaotic mapping, it can be represented as: ; In the formula, For the first The chaotic value of the next iteration; To control the key parameters of the segmented intervals; The mapping is in The interval is divided into four linear interval transformations (control parameters). (After experimental verification, the distribution is optimal), the generated chaotic sequence has stronger ergodicity and more uniform distribution; the chaotic sequence is linearly mapped to the coordinate range of the path points defined by S1 to obtain a fully covered initial population. In each generation of the population, the bottom 30% of individuals by fitness are selected, based on the current population dynamic boundary (minimum values of each dimension). Maximum value Nonlinear scaling factor with chaotic sequence control (with iterations from) Smooth transition to ), calculate the inverse solution group; the formula for calculating the inverse solution is expressed as: ; ; ; ; In the formula, For the first Inverse solution coordinates in each dimension; For the first The first generation of the population The minimum value of the dimension; This is a non-linear scaling factor; For the first The first generation of the population The maximum value of the dimension; For individuals in the first Dimensional coordinates; This represents the total number of iterations. This represents the current iteration number; For the first In the first generation of the population, all individuals in the second generation... A set of coordinates of a dimension; This is the minimum value of the scaling factor; This represents the maximum value of the scaling factor. By employing elite retention and greedy selection, high-quality reverse solutions are incorporated into the population, thus expanding the scope of solution space exploration. Real-time monitoring of population diversity indicators (average distance between individuals, fitness variance) is performed. When diversity falls below a threshold, a chaotic sequence is generated as a perturbation factor and injected into individual location updates. The perturbation formula is expressed as: ; In the formula, For the original position of the individual, The position after the disturbance. For legal operations within the solution space, These are chaotic random numbers generated by piecewise chaotic mapping. The adaptive perturbation strength coefficient is related to the number of iterations. This represents the search scope scale for the current dimension; S3 receives the constraint model from S1 and the population from S2, and simulates the dynamic role division of the sparrow population, including discoverers (global exploration), followers (local mining), and vigilants (risk monitoring); it calculates the fitness of each individual in the population based on the constraint model from S1, drives the population to iteratively seek optimization, and balances global exploration and local optimization; it outputs the population dynamic data and convergence status indicators during the iteration process. In this step, individuals are ranked based on their fitness values calculated using S1, and roles are assigned proportionally: Discoverers (top 20%): These have the best fitness and are responsible for global exploration. Their positions are updated using nonlinear inertial weights and adaptive step sizes, represented as follows: ; ; In the formula, Nonlinear dynamic weights; This is an adaptive step size factor; For the first In the first generation of the population, the first The individual in the first The position coordinates of the dimension; For the first In the first generation of the population, the first The individual in the first The original position coordinates of the dimension; For population alert status parameters; This is a safety threshold; A normally distributed random factor; For the first The optimal individual in the population in the 1st generation The position coordinates of the dimension; These are random numbers distributed according to a standard normal distribution. This is the step size control factor; This represents the minimum value of the inertia weight. This represents the maximum value of the inertia weight; It is a nonlinear decay term; By combining the group's historical best information, convergence to high-quality regions is accelerated; Followers (middle 60%-70%): Responsible for local fine-tuning, integrating and improving sine and cosine algorithms, switching between exploration / mining modes according to crossover probability, expanding the search range during exploration, and focusing on local optimization during mining, represented as: ; In the formula, , For dynamic control parameters; These are uniformly distributed random numbers in the interval 0-1. For exploration factors; For development factors; The crossover probability; For the first The position coordinates of an individual in the population; For the first The original location coordinates of an individual in the population; For the first The coordinates of a high-quality individual (such as the discoverer or the current best individual) in the population; The weighted difference modulus between the superior individual and the current follower's position; Vigilant Group (Last 10%-20%): Establish an adaptive early warning mechanism based on risk perception for risk monitoring; if an individual's fitness is worse than the population mean, it migrates to the optimal region; if the population aggregation is too high, it triggers local diffusion to avoid premature convergence, represented as: ; In the formula, The fitness tolerance threshold, For the disturbance intensity, This is the clustering threshold. For random disturbance factors; This represents the fitness value of an individual. This represents the fitness value of a reference individual in the population. This is the local search step size coefficient; This represents the fitness value of the worst-performing individual in the population. For individual crowding levels; In each iteration, each role updates its position based on its own strategy and calculates a new fitness through the comprehensive cost function of S1; individuals dynamically adjust their strategies according to changes in fitness to form a "competition-cooperation" equilibrium; at the same time, chaotic perturbations of S2 are incorporated to break the local convergence deadlock. S4 receives population distribution data, fitness evolution curves, and convergence status indicators from S3, and integrates the convergence monitoring unit for real-time analysis; it feeds back the adjusted parameter configurations (inertia weight values, perturbation intensity, and crossover probability) to the optimized chaotic perturbation logic of S2 and the optimized role update strategy of S3, forming a closed-loop control. In this step, if a decrease in search efficiency is detected, a parameter reorganization mechanism is automatically triggered: Adjust the inertia weight of the discoverer in S3, the crossover probability of the follower, and the intensity coefficient of the chaotic perturbation in S2; for example, increase the weight of the discoverer in the early stage of the iteration to enhance exploration, and increase the mining factor of the follower in the later stage to accelerate convergence. When switching scenes (such as from open farmland to urban high-rise areas), parameters can be adaptively calibrated without manual intervention, ensuring that optimization efficiency fluctuations do not exceed 5%. S5 receives the optimal individual data converged by S3 and the constraint model data of S1, and performs feasibility verification on the candidate paths based on the physical parameter constraints and obstacle avoidance rules of S1; after the verification is passed, it outputs the optimal path of the UAV that takes into account flight safety, kinematic feasibility and mission requirements. In this step, the feasibility verification includes: Physical constraints: Check if the total flight distance is less than or equal to the maximum flight distance, and if the altitudes of all waypoints are within the specified limits. Are the inner turning angle ≤ the maximum turning angle? Safety constraints: Check that the path segment is far from all cylindrical threat zones; Kinematic feasibility: Check path smoothness; Recalculate the comprehensive cost of candidate paths It was confirmed that it met the triple requirements of safety, feasibility, and economy; After successful verification, the complete optimal path data is output for the UAV flight control system to use for autonomous navigation.
[0021] Therefore, this invention adopts the above-mentioned UAV path planning method and system based on chaotic multi-agent game optimization. This method enhances the global exploration capability through segmented chaotic mapping, lens imaging reverse learning, and chaotic disturbance monitoring; constructs constraint models and hard constraint physical parameters and optimizes obstacle modeling to improve local accuracy; optimizes multi-objective balance by using dynamic weights and three types of constraints; simulates sparrow division of labor to speed up, adds monitoring units to stabilize efficiency, and comprehensively optimizes the UAV path planning performance.
[0022] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0023] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.
Claims
1. A method for unmanned aerial vehicle path planning based on chaotic multi-agent game optimization, characterized in that, The method comprises the following steps: S1, constructing a three-dimensional comprehensive constraint model comprising path length, flight height, and path smoothness, setting hard constraints for the maximum turning angle of the unmanned aerial vehicle and the physical parameters of the flight height corridor; modeling the obstacles as cylindrical threat zones, and introducing a penalty term related to the intrusion depth, outputting the constraint model, the physical parameter constraint threshold, the obstacle modeling parameters, and the intrusion depth penalty logic; S2, generating an initial population by segmented chaotic mapping, and generating a reverse solution population by lens imaging reverse learning; Real-time monitoring of population fitness variance, injecting chaotic disturbance when diversity is below the threshold, and maintaining chaotic state stability, outputting the initial population, chaotic disturbance triggering threshold, and chaotic state monitoring parameters; S3, receiving the constraint model of S1 and the initial population of S2, simulating the dynamic role division of finders, followers, and alarmers in sparrow populations, calculating the fitness of population individuals based on the constraint model of S1 and driving population iterative optimization, outputting population dynamic data, convergence state indicators, and optimal individual data of each generation; S4, receiving the iterative data of S3, real-time analyzing population distribution and fitness evolution curve through a convergence monitoring unit, dynamically adjusting inertia weight, disturbance intensity, and crossover probability, and feeding back the adjusted parameters to S2 and S3 to form a closed-loop regulation; S5, receiving the optimal individual data of S3 and the constraint model of S1, performing physical constraint, safety constraint, and kinematic feasibility verification on the candidate path, and outputting the unmanned aerial vehicle path after verification.
2. The method of claim 1, wherein, In S1, the expression of the three-dimensional comprehensive constraint model is: ； wherein, is a comprehensive cost; is a path length cost; is a flight height cost; is a path smoothness cost; , and are positive weight coefficients of each cost, and satisfy .
3. The method of claim 1, wherein, In S1, the path length cost is the sum of the three-dimensional Euclidean distances of all consecutive path points, and the total distance satisfies ； In the formula, is the maximum range of the drone; is the total range; The flight height cost is the standard deviation of the height of the path point, expressed as: ； ； In the formula, This represents the average flight altitude along the entire route. For the first The flight altitude corresponding to each waypoint; and each path point height is full ； wherein is the minimum flight height allowed for the drone; is the maximum flight height allowed for the drone; The path smoothness cost includes a turn angle over-penalty term and a climb angle variation penalty term, expressed as ； In the formula, The weighting coefficient for excessive penalty for turning angle; For the first The climb angle corresponding to each path point; This represents the absolute change in the climb angle between adjacent path segments; The weighting coefficient for the change in the climb angle; This is an excess penalty for turning angles; This represents the total number of path points. The index is the sequence number of the path point; The intrusion depth penalty term expression is: ； wherein is a penalty factor; is an obstacle threat cost of the path; is a threat radius; is a closest distance of the threat center.
4. The method of claim 1, wherein, In S2, the expression of the segmented chaotic mapping is: ； In the formula, is the first chaotic value of the first iteration; is a key parameter for controlling the segment interval, .
5. The method of claim 1, wherein, In S2, the reverse solution calculation formula of the lens imaging reverse learning is: ； ； ； ； In the formula, For the first Inverse solution coordinates in each dimension; For the first The first generation of the population The minimum value of the dimension; This is a non-linear scaling factor; For the first The first generation of the population The maximum value of the dimension; For individuals in the first Dimensional coordinates; This represents the total number of iterations. This represents the current iteration number; For the first In the first generation of the population, all individuals in the second generation... A set of coordinates of a dimension; This is the minimum value of the scaling factor; This represents the maximum value of the scaling factor. The formula of the chaotic disturbance is: ； wherein, is the original position of the individual, is the position after perturbation, is a legal operation within the solution space, is a chaotic random number generated by a piecewise chaotic mapping, is an adaptive perturbation strength coefficient related to the iteration number, is the search range scale of the current dimension.
6. The method of claim 1, wherein, In S3, the proportion of finders in the population is 20%, the position updating strategy of finders contains safe and unsafe scenarios, and the inertia weight expression is: ； ； wherein, is a non-linear dynamic weight; is an adaptive step factor; is the position coordinate of the i-th individual in the j-th dimension in the population of the g-th generation; is the position coordinate of the i-th individual in the j-th dimension in the population of the g-th generation; is the original position coordinate of the i-th individual in the j-th dimension in the population of the g-th generation; is the original position coordinate of the i-th individual in the j-th dimension in the population of the g-th generation; is the original position coordinate of the i-th individual in the j-th dimension in the population of the g-th generation; is the original position coordinate of the i-th individual in the j-th dimension in the population of the g-th generation; is the original position coordinate of the i-th individual in the j-th dimension in the population of the g-th generation; is the original position coordinate of the i-th individual in the j-th dimension in the population of the g-th generation; is the population alertness state parameter; is the safety threshold; is the normal distribution random factor; is the position coordinate of the optimal individual in the j-th dimension in the population of the g-th generation; is the position coordinate of the optimal individual in the j-th dimension in the population of the g-th generation; is the position coordinate of the optimal individual in the j-th dimension in the population of the g-th generation; is the standard normal distribution random number; is the step control factor; is the minimum value of the inertia weight; is the maximum value of the inertia weight; is the non-linear decay term; The follower occupies 60-70% of the population, and its position update strategy is based on a random number The size of the cross probability Switches between exploration and exploitation modes, expressed as: ； wherein , is a dynamic control parameter; is a uniformly distributed random number in the interval 0-1; is an exploration factor; is an exploitation factor; is a crossover probability; is the position coordinate of an individual in the population of generation ; is the original position coordinate of an individual in the population of generation ; is the position coordinate of a good individual (e.g. the discoverer, the current best individual) in the population of generation ; is the weighted difference modulus between the position of the good individual and the current follower. The proportion of alarmers in the population is 10%-20%, and the position updating strategy of alarmers uses different adjustment methods according to the difference between individual fitness and reference individual fitness and the population crowding degree, expressed as: ； wherein, is a fitness tolerance threshold, is a perturbation strength, is a degree of convergence threshold, is a random perturbation factor; is a fitness value of an individual; is a fitness value of a reference individual in the population; is a local search step length coefficient; is a fitness value of the worst individual in the population; is a crowding degree of an individual.
7. The method of claim 1, wherein, In S4, the convergence monitoring unit judges the search efficiency by analyzing population aggregation, optimal fitness growth rate, and search trajectory, and when the efficiency is detected to be decreased, the inertia weight of finders is increased in the early stage of iteration to strengthen global exploration, and the exploitation factor of followers is improved in the later stage of iteration to speed up local convergence.
8. The unmanned aerial vehicle path planning system based on chaotic multi-agent game optimization, applied to the unmanned aerial vehicle path planning method based on chaotic multi-agent game optimization according to any one of claims 1-7, characterized in that, It comprises: A constraint model construction module for executing S1, constructing a three-dimensional comprehensive constraint model, modeling the cylindrical threat zones of obstacles, and setting intrusion depth penalty rules, and outputting corresponding evaluation and constraint parameters; A population initialization and diversity maintenance module connected with the constraint model construction module for executing S2, generating an initial population and a reverse solution population by segmented chaotic mapping and lens imaging reverse learning, realizing chaotic disturbance and state monitoring, and outputting related population data; A multi-agent game iteration module, connected with the population initialization and diversity maintenance module, is configured to execute S3 to realize multi-role division of labor iteration optimization of the sparrow population and output iteration data. A parameter dynamic correction module, connected with the multi-agent game iteration module, is configured to execute S4 to complete convergence state analysis and parameter closed-loop regulation. A path verification output module, connected with the constraint model construction module and the multi-agent game iteration module, is configured to execute S5 to complete multi-dimensional path verification and output an optimal path.
9. A computer device, comprising: The processor is configured to be coupled with the memory, read and execute instructions and / or program codes in the memory, and execute the method in any one of claims 1-7. The computer readable medium stores computer program codes, which, when running on a computer, cause the computer to execute the method in any one of claims 1-7.
10. A computer readable medium characterized by