Method, device and equipment for planning air-ground cooperative task of intelligent agent, and medium

Abstract

The invention provides a method, a device, equipment and a medium for planning an air-ground cooperative task of an intelligent agent, and the method forms a closed-loop technical route of task modeling, communication and information consistency, cost and distribution optimization and trajectory control, and after task distribution and trajectory planning are completed, multiple unmanned aerial vehicles and multiple unmanned vehicles are in full communication under the condition of complete communication. The intelligent agents can acquire global task and state information in real time, the planned trajectory is smooth and feasible, enough safe spacing is kept between the intelligent agents, dynamic target tracking and regional patrol tasks are completed in sequence, and cooperation of task allocation and trajectory planning is achieved. Under the condition of incomplete communication when local information interaction is carried out in a communication radius, all agents can still gradually form consistent task cost cognition by means of a consistency algorithm, cooperative scheduling and trajectory tracking are completed on the basis of local information, and the overall task completion quality and the cooperative effect are basically kept consistent with a complete communication scene.

Description

Technical Field

[0001] This invention belongs to the technical field of air-ground collaborative mission planning, specifically relating to a method, apparatus, equipment, and medium for planning air-ground collaborative missions of intelligent agents. Background Technology

[0002] With the rapid development of unmanned systems technology, utilizing unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs) to collaboratively complete complex tasks has become an important trend in intelligent transportation and intelligent security. UAVs have advantages such as high mobility, wide field of view, and easy acquisition of global information; UGVs, on the other hand, have stable endurance and large payload capacity in ground environments. When the two cooperate, they can leverage their respective advantages to complete tasks such as dynamic tracking, area surveillance, and rescue patrols. However, due to the different kinematic models of UAVs and UGVs, the complexity and variability of mission objectives, and the constraints of time windows, achieving efficient task allocation and trajectory control while meeting requirements such as time, energy consumption, and path safety remains a challenge in current research.

[0003] The Whale Optimization Algorithm (WOA) is primarily suitable for continuous optimization problems. In discrete task allocation and multi-objective optimization scenarios, its convergence factor is fixed and the search range lacks dynamic adjustment, making it prone to premature convergence. In recent years, researchers have improved the search performance of WOA by modifying the convergence factor, introducing chaotic mappings, or Gaussian perturbations. However, these improvements rarely incorporate prior information about task allocation, resulting in low computational efficiency.

[0004] In terms of control, since drones and unmanned vehicles have different dynamic constraints, traditional PID control or control strategies based on linear quadratic regulators are difficult to simultaneously consider input constraints, obstacle avoidance, and time windows in dynamic environments. Model Predictive Control (MPC), by predicting the future behavior of the system within a finite time range and solving for the optimal control sequence, can handle the dynamic constraints and control input limitations of the system well and has good robustness, and is considered an effective method for solving multi-agent path planning and control. However, the performance of MPC depends on the task allocation result; if the allocation scheme is unreasonable, even the best control optimization will struggle to achieve overall optimality.

[0005] Therefore, this application anticipates a method for air-ground collaborative task planning based on a greedy auction algorithm and an improved whale optimization algorithm. Summary of the Invention

[0006] To overcome the problems of low task allocation efficiency, easy getting trapped in local optima, and difficulty in coordinating trajectory control of different platforms in the existing technology when handling air-ground cooperative tasks, the present invention provides a method, apparatus, device and medium for planning intelligent agent air-ground cooperative tasks.

[0007] To achieve the aforementioned objective, the present invention provides the following technical solution:

[0008] In a first aspect, embodiments of this disclosure provide a method for planning intelligent agent air-ground cooperative tasks, comprising the following steps:

[0009] Step S1: The intelligent agent includes drones and unmanned vehicles. Construct the task scenario of the intelligent agent and the two-dimensional kinematic model of the drone and the two-dimensional kinematic model of the unmanned vehicle.

[0010] Step S2: Construct the complete communication information interaction structure and the incomplete communication information interaction structure of the intelligent agent;

[0011] Step S3: Based on the intelligent agent task scenario, the two-dimensional kinematic model of the UAV, and the two-dimensional kinematic model of the unmanned vehicle, construct the cost model and comprehensive objective function of the intelligent agent in an integral form, and obtain the local comprehensive cost;

[0012] Step S4: Based on the local comprehensive cost and introducing the task urgency factor, obtain the bidding results of each agent for each task. Based on the bidding results, adopt the pre-allocation mechanism of the greedy auction algorithm, traverse the task scenarios under the complete communication information interaction structure and the incomplete communication information interaction structure respectively, and obtain the task pre-allocation matrix and the corresponding task pre-allocation quality index.

[0013] Step S5: The improved whale optimization algorithm is used to optimize the initial allocation scheme according to the task pre-allocation quality index to obtain the optimal task allocation matrix. The initial allocation scheme is the task pre-allocation matrix.

[0014] Step S6: Construct the UAV prediction model and the unmanned vehicle prediction model, as well as the intelligent agent's reference output trajectory and reference control sequence, and combine them with the optimal task allocation matrix to obtain the final planning result.

[0015] Furthermore, the task scenarios in step S1 include scenarios of various scales and tasks of various types. The scenarios of various scales include small-scale scenarios, medium-scale scenarios and large-scale scenarios. The task types include dynamic moving target tracking and fixed area patrol and surveillance.

[0016] The task set in the task scenario Represented as: , Indicates the first One task;

[0017] In a two-dimensional task space, the task The starting point and the ending point are represented as follows:

[0018] ;

[0019] In the formula, Let be the two-dimensional coordinates at the start of task j; Let task j start with its horizontal coordinate value; The vertical coordinate value at the start of task j; The two-dimensional coordinates at the end of task j; The horizontal coordinate value at the end of task j; This represents the vertical coordinate value at the end of task j;

[0020] Configure constraints with deadlines. And define a binary task assignment matrix. Represented as:

[0021] ;

[0022] In the formula, Indicates the first One intelligent agent; Represents intelligent agents Assigned to perform tasks , This indicates that the task has not been assigned. Task assignment satisfies the following assignment constraints:

[0023] ;

[0024] ;

[0025] In the formula, For a collection of intelligent agents, The assignment constraint is used to configure each task to be executed by an agent, and each agent executes only one task within the same planning period.

[0026] In step S1, the motion state of the agent in the two-dimensional scene is represented by its planar position and heading angle. planar position for:

[0027] ;

[0028] In the formula, Let be the horizontal coordinate value of agent i; Here is the vertical coordinate value of agent i;

[0029] The drone flies at a constant altitude in the air, considering only the drone's position... For translation and heading changes within a plane, the two-dimensional kinematic model of the UAV includes:

[0030] ;

[0031] In the formula, Let be the velocity of the drone in the horizontal x-axis direction; Let be the velocity of the drone in the vertical y-axis direction; The angular velocity of the drone's heading; This refers to level flight speed; For heading angle; This is the yaw angle control input; , For a subset of drones, ;

[0032] The unmanned vehicle moves in the ground plane, and the constraint relationship between velocity and steering is described using the Ackermann kinematic model, resulting in a two-dimensional kinematic model of the unmanned vehicle including:

[0033] ;

[0034] In the formula, Let x be the velocity of the autonomous vehicle in the horizontal x-axis direction; Let be the velocity of the autonomous vehicle in the vertical y-axis direction; Let be the angular velocity of the autonomous vehicle's heading. For the vehicle's heading angle, For vehicle wheelbase, The driving speed of the driverless car, This refers to the front wheel steering angle; For driverless car subsets, .

[0035] Furthermore, under the fully communicative information interaction structure in step S2, all agents share global information, and the neighborhood set of an agent is represented as:

[0036] ;

[0037] Under the aforementioned incomplete communication information interaction structure, agents can only communicate with other agents in spatial proximity, within a communication radius. The neighborhood set is defined as follows:

[0038] ;

[0039] In the formula, For agents within the neighborhood of agent i under incomplete communication; The position of agent i; Let j be the position of agent j.

[0040] Furthermore, the cost model in step S3 includes:

[0041] ;

[0042] ;

[0043] In the formula, and These are the speed and steering inputs for the unmanned vehicle, or the yaw control inputs for the drone; These are the weighting coefficients; For intelligent agents The planar position vector at the current moment; For the task The starting position vector; For the task The endpoint position vector; For intelligent agents The final stopping position vector;

[0044] The integrated objective function includes:

[0045] ;

[0046] In the formula, The total time to complete the task; Task deadline; weight The importance of adjusting the path, controlling energy consumption, and penalties for overdue payments; Assign matrix to task The Middle The first agent and the second Binary decision variables between tasks; For intelligent agents Execute the task The path cost at any given time represents the distance the agent travels from its current position to the task. The sum of the path length or distance from the starting point to the destination after completing the task; For intelligent agents Execute the task Cost control.

[0047] Furthermore, the bidding results in step S4 include:

[0048] ;

[0049] in It is a regular term; For local comprehensive costs; As a task urgency factor, ; For the current moment, It is a constant, and Task deadline The closer they are, the higher the task urgency factor. The larger;

[0050] For the same task Local comprehensive cost The smaller the bid, the better. The larger the urgency factor, the tighter the deadline for different tasks. The larger the size, the higher the priority it will be in the auction;

[0051] A greedy auction algorithm is used for pre-allocation. For the fully communicative information interaction structure, the bidding results of the agents for the corresponding tasks are auctioned across the entire network. For the incompletely communicative information interaction structure, local bidding is conducted in the neighborhood set. After the local bidding ends, each agent broadcasts its current temporary task allocation result and corresponding cost information to its neighbors in multiple communication rounds. Based on the received neighbor information, the agent updates the local task allocation variables and resolves conflicts until the decisions of each agent in the neighborhood regarding task ownership no longer change, resulting in a consistent task pre-allocation matrix across the entire network.

[0052] The bidding process includes:

[0053] Step A1: Initialize the set of unassigned tasks A collection of intelligent agents is available. ;

[0054] Step A2: In all Find the highest bid result:

[0055] ;

[0056] In the formula, For tasks in the unassigned task set; Agents are those in the set of available agents;

[0057] In the context of a fully interactive communication structure, A represents the set of all intelligent agents; in the context of a partially interactive communication structure, A represents the set of intelligent agents participating in the auction within the current communication neighborhood.

[0058] Step A3: Based on the maximum bid result, allocate tasks from the unassigned task set. Agents assigned to the set of available agents ,set up and from the set of available agents and unassigned task set Remove the corresponding agent or task from the list;

[0059] Step A4: Repeat steps A2–A3 until all tasks have been assigned or no agents are available, to obtain the initial assignment scheme;

[0060] Step A5: Based on the pre-allocation results and local comprehensive costs Obtain the task pre-assignment matrix and task pre-assignment matrix The corresponding global objective function value for:

[0061] ;

[0062] In the formula, Pre-allocate matrix for tasks The Middle Line 1 Column elements, It is a binary decision variable, representing the agent. With the task In the pre-allocation phase, the association relationship is such that a value of 1 indicates allocation, and a value of 0 indicates no allocation.

[0063] Step A6: Based on the global objective function value Construct task pre-allocation quality indicators :

[0064] ;

[0065] In the formula, These represent the empirical optimal and empirical worst values ​​in the feasible solution space under a given task scenario, respectively. Based on the empirical optimal value... With worst experience It characterizes the relative merits of the current initial solution.

[0066] Furthermore, in step S5, an adaptive nonlinear convergence factor, a Gaussian perturbation local exploitation strategy, and an optimal solution replacement strategy are introduced into the whale algorithm to obtain an improved whale optimization algorithm, which is then based on task pre-allocation quality indicators. The optimal task allocation matrix is ​​obtained by globally optimizing the task pre-assignment matrix. ;

[0067] The adaptive nonlinear convergence factor Based on task pre-allocation quality indicators Adjusting the convergence factor exponent includes the following steps:

[0068] ;

[0069] In the formula, To adjust the parameters; This represents the maximum number of iterations. for To be based on pre-allocated quality indicators; A changing power function; This is the initial value of the exponent;

[0070] The local exploitation strategy for the Gaussian perturbation includes:

[0071] ;

[0072] In the formula, , The current iteration algebra The position vector of the individual with the lowest fitness in the next population is used as the position corresponding to the global optimal solution in the current iteration; For the first The era The position vector of the nth individual, corresponding to the nth individual Encoded representation of each candidate task allocation scheme; It is a spiral contraction factor; The helical coefficient is... Randomly generated, Controlling noise amplitude Let be a standard Gaussian random variable with a mean of 0 and a variance of 1;

[0073] The optimal solution replacement strategy includes: replacing the worst solution with the current optimal solution after each iteration. ,but ;

[0074] In the formula, This is the worst solution; This is the optimal solution;

[0075] The improved whale optimization algorithm is based on parameters Determine whether to perform a bounding or searching operation, and then update the entire system accordingly.

[0076] ;

[0077] In the formula, In order to be in Uniformly distributed random numbers over an interval For coefficient vectors;

[0078] like Execute the encirclement strategy;

[0079] ;

[0080] like Randomly select individuals search

[0081] ;

[0082] In the formula, This indicates element-wise multiplication. Individuals randomly selected from the current population; For the first Era Intelligent Agent The position vector, and the corresponding encoded representation of the candidate task allocation scheme;

[0083] When combining three improvement strategies to obtain the optimal task allocation matrix, the following are included:

[0084] Step B1: Assign the task pre-matrix Mapped to individual whales according to a preset encoding method. and in the task pre-assignment matrix Several individuals are generated within the neighborhood of the whale, forming the initial population for the improved whale optimization algorithm; each individual... Each corresponds to a candidate task assignment matrix;

[0085] Step B2: For any iterative algebra Individuals in Decoding yields the corresponding current task allocation matrix. And calculate the fitness based on the comprehensive objective function:

[0086] ;

[0087] in, Using the comprehensive objective function defined in step S3, sort the individuals by fitness values ​​from smallest to largest to obtain the current optimal individual. and its corresponding task allocation matrix and the worst individual at present ;

[0088] Step B3: In the In the iteration, the given adaptive nonlinear convergence factor is first used. Calculate the parameter vector:

[0089] ;

[0090] In the formula, In order to be in Uniformly distributed random numbers over an interval For coefficient vectors;

[0091] when At this time, corresponding to the encirclement and local development stages: for individuals closest to the current optimal individual individual First calculate Based on the standard bounding update, a local exploitation strategy for the defined Gaussian perturbation is introduced to update the individual positions:

[0092] ;

[0093] In the formula, , It is a spiral contraction factor; The helical coefficient; The number is randomly generated; These are noise amplitude control parameters; Let be a standard Gaussian random variable with a mean of 0 and a variance of 1; by superimposing Gaussian perturbations in the enclosing operation, the search for the neighborhood of the current optimal task assignment matrix is ​​achieved;

[0094] when At this time, corresponding to the global search phase: randomly select individuals. As a search reference, for individuals Update:

[0095] ;

[0096] In the formula, This indicates element-wise multiplication. Individuals randomly selected from the current population are used to expand the global exploration range of the solution space;

[0097] Step B4: Complete for all individuals After the position is updated, the fitness of the new population is recalculated to obtain the best individuals of the new generation. With the worst individual According to the given optimal solution replacement strategy, if Then perform the replacement operation. Ensure that at least one currently optimal task allocation scheme is retained in each generation of the population;

[0098] Step B5: Repeat steps B2 to B4 until the preset maximum number of iterations is reached. Termination conditions include fitness changes falling below a preset threshold.

[0099] At termination, the individual with the lowest fitness is selected from the final population. and will Decoding yields the corresponding optimal task allocation matrix. The optimal task allocation matrix This represents the optimal task allocation result.

[0100] Furthermore, in step S6, the drone prediction model includes:

[0101] A state-space-based discrete-time prediction model is used to model the UAV, and the UAV prediction model includes:

[0102] ;

[0103] In the formula, Index for drones; For drones at discrete time The state vector; This is the control input vector for the UAV. This is the output vector of the UAV; Used to select the UAV's planar position from the state vector as the output; This is the discrete-time system matrix of the UAV, which is based on the UAV's two-dimensional kinematic model after sampling period. Linearization and discretization yield:

[0104] The continuous-time nonlinear model can be written as:

[0105] ;

[0106] In the formula, ;

[0107] In the reference trajectory For nonlinear functions Linearization is performed using a first-order Taylor expansion, and zero-order preservation or Euler approximation is used according to the sampling period. Discretize the system and obtain the discrete-time system matrix according to linear system theory:

[0108] ;

[0109] In the formula, , For the selected linearized operating point, It is a third-order identity matrix; , These are the Jacobian matrices for the state and control, respectively;

[0110] The autonomous vehicle prediction model includes:

[0111] The autonomous vehicle prediction model is constructed using a state-space-based discrete-time prediction model, including:

[0112] ;

[0113] in, Index for driverless cars; Used to select the position of the unmanned vehicle in the plane as the output from the state vector; For autonomous vehicles at discrete times The state vector; This is the control input vector for the autonomous vehicle; This is the output vector of the autonomous vehicle; This is the discrete-time system matrix of the autonomous vehicle, which is based on the two-dimensional kinematic model of the autonomous vehicle after sampling period. Linearization and discretization yield:

[0114] The continuous-time nonlinear model can be written as:

[0115] ;

[0116] In the formula, L is the vehicle wheelbase;

[0117] In the reference trajectory For nonlinear functions Linearization is performed using a first-order Taylor expansion, and zero-order preservation or Euler approximation is used according to the sampling period. Discretize the system and, according to linear system theory, obtain the discrete-time system matrix:

[0118] ;

[0119] In the formula, For the selected linearized operating point, It is a third-order identity matrix. , These are the Jacobian matrices for the state and control, respectively;

[0120] The agent's reference output trajectory includes:

[0121] ;

[0122] In the formula, Indicates the current control cycle Starting from the predicted step size The desired planar position reference trajectory point at any given time. To predict the length of the time domain;

[0123] The reference control sequence includes:

[0124] ;

[0125] In the formula, This indicates the prediction step size in order to achieve the reference output trajectory. The desired control input reference value to be applied at any given time; where and Collectively referred to as , and Collectively referred to as ;

[0126] The process of obtaining the path planning result includes:

[0127] In each control cycle Inside, in the current state Using the initial conditions and the aforementioned UAV prediction model or unmanned vehicle prediction model as constraints, a finite-time optimization problem is constructed:

[0128] ;

[0129] When the state constraints are satisfied Control constraints Under the given conditions, the optimal control sequence is obtained. and the first control variable in the optimal control sequence The path planning result of the current control cycle is output and applied to the actual system, and the state at the next moment is updated by the UAV prediction model or the unmanned vehicle prediction model. ;

[0130] The entire task execution process includes Each control cycle, through Repeat steps C1 to C3 within each control cycle to obtain the complete control quantity sequence and path planning results. The specific steps are as follows:

[0131] Step C1: In the control cycle At the beginning, the state is measured or estimated at the current time. As initial conditions, the finite-time optimization problem is constructed and solved to obtain the optimal control sequence in the prediction time domain. ;

[0132] Step C2: Use the first control variable in the optimal control sequence as the actual control input for the current cycle:

[0133] ;

[0134] And Send to the execution layer of the drone or unmanned vehicle, and record simultaneously. and the resulting state As part of the path planning results;

[0135] Step C3: Update the system state using the discrete-time prediction model to obtain the state at the next sampling time:

[0136] ;

[0137] And order Enter the next control cycle and repeat steps C1 to C3 until... The control input sequence is obtained sequentially:

[0138] ;

[0139] The control input sequence is input into the UAV prediction model or the unmanned vehicle prediction model for recursion to obtain the corresponding state sequence. And by the output equation The calculated output trajectory The output trajectory is used as the complete path planning result from the start time of the task to the end time of the task.

[0140] Summarize all intelligent agents The output trajectory is used to obtain the final path planning result of the multi-agent system in the entire task time domain and is used as the final planning result.

[0141] In a third aspect, embodiments of this disclosure provide a system for planning intelligent agent air-ground cooperative tasks, comprising:

[0142] The initialization unit is configured to: construct the task scenario of the intelligent agent, including drones and unmanned vehicles, and the two-dimensional kinematic model of the drones and the two-dimensional kinematic model of the unmanned vehicles;

[0143] The communication unit is configured to: construct a complete communication information interaction structure and a non-complete communication information interaction structure for the intelligent agent;

[0144] The cost unit is configured as follows: based on the agent's task scenario, the UAV's two-dimensional kinematic model, and the unmanned vehicle's two-dimensional kinematic model, the cost model and comprehensive objective function of the agent are constructed in an integral form, and the local comprehensive cost is obtained.

[0145] The pre-allocation unit is configured to: obtain the bidding results of each agent for each task based on the local comprehensive cost and by introducing a task urgency factor; and, based on the bidding results, adopt a pre-allocation mechanism using a greedy auction algorithm to traverse the task scenarios under both fully communicative information interaction structures and incompletely communicative information interaction structures to obtain a task pre-allocation matrix and corresponding task pre-allocation quality indicators.

[0146] The optimal allocation unit is configured to: use an improved whale optimization algorithm to optimize the initial allocation scheme according to the task pre-allocation quality index to obtain the optimal task allocation matrix, wherein the initial allocation scheme is the task pre-allocation matrix;

[0147] The output unit is configured to: construct a UAV prediction model and an unmanned vehicle prediction model, as well as an agent reference output trajectory and a reference control sequence, and combine them with the optimal task allocation matrix to obtain the final planning result.

[0148] In a third aspect, embodiments of this disclosure provide an electronic device, characterized in that the electronic device comprises:

[0149] At least one processor; and,

[0150] The memory is communicatively connected to the at least one processor; wherein,

[0151] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method for planning intelligent agent air-ground collaborative tasks.

[0152] In a fourth aspect, embodiments of this disclosure provide a non-transitory computer-readable storage medium, characterized in that the non-transitory computer-readable storage medium stores computer instructions for causing the computer to perform the method for planning intelligent agent air-ground cooperative tasks.

[0153] The beneficial effects of this invention are:

[0154] This invention provides a method, apparatus, device, and medium for planning intelligent agent air-ground collaborative tasks. This method, through the construction of task scenarios, communication structures, and task allocation, achieves safe, collaborative, and energy-efficient execution trajectories for multiple UAVs and multiple unmanned vehicles in a two-dimensional task space, forming a closed-loop technical route encompassing task modeling, communication and information consistency, cost and allocation optimization, and trajectory control. After completing task allocation and trajectory planning, under full communication conditions, each intelligent agent can obtain global task and status information in real time. The planned trajectory is smooth and feasible, with sufficient safe distances maintained between agents. Simultaneously, dynamic target tracking and area patrol tasks are completed sequentially, achieving collaboration between task allocation and trajectory planning. Under incomplete communication conditions where communication is limited and only local information exchange is allowed within the communication radius, each intelligent agent, relying on a consensus algorithm, can still gradually form a consistent understanding of task costs and complete collaborative scheduling and trajectory tracking based on local information. The overall task completion quality and collaborative effect are essentially consistent with those in fully communicative scenarios. This method can effectively achieve collaborative task execution of UAVs and unmanned vehicles under time, energy consumption, and path safety constraints, and possesses global search capabilities and stable trajectory control performance. Attached Figure Description

[0155] Figure 1 A flowchart illustrating a method for planning an intelligent agent's air-ground cooperative task according to an embodiment of this disclosure is shown;

[0156] Figure 2 A schematic diagram of an air-ground cooperative mission scenario according to an embodiment of the present disclosure is shown, including multiple drones and unmanned vehicles as well as a mission target;

[0157] Figure 3 A flowchart of the Greedy Auction Algorithm (GAAI) according to an embodiment of this disclosure is shown;

[0158] Figure 4 A schematic diagram of the greedy auction algorithm combined with the improved whale optimization algorithm (GAAI-IWOA) according to an embodiment of the present disclosure is shown;

[0159] Figure 5 A block diagram of the model predictive control structure according to an embodiment of the present disclosure is shown;

[0160] Figure 6 The diagram shows a dynamic rendering of GAAI-IWOA under fully communicative conditions according to an embodiment of this disclosure.

[0161] Figure 7 The diagram illustrates a cost comparison of different algorithms in a small-scale scenario under fully communicative conditions, including control cost, path cost, and total cost, according to embodiments of this disclosure.

[0162] Figure 8 The diagram shows a comparison of algorithm performance metrics in a small-scale scenario under fully communicative conditions, including a comparison of algorithm runtime, repetition rate, and error rate.

[0163] Figure 9 The diagram illustrates a cost comparison of different algorithms in a medium-scale scenario under fully communicative conditions, including control cost, path cost, and total cost, according to embodiments of this disclosure.

[0164] Figure 10 The diagram shows a comparison of algorithm performance metrics in a medium-scale scenario under fully communicative conditions, including a comparison of algorithm runtime, repetition rate, and error rate.

[0165] Figure 11 The diagram illustrates a cost comparison of different algorithms in a large-scale scenario under fully communicative conditions, including control cost, path cost, and total cost, according to embodiments of this disclosure.

[0166] Figure 12 The diagram illustrates a comparison of algorithm performance metrics in a large-scale scenario under fully communicative conditions, including a comparison of algorithm runtime, repetition rate, and error rate.

[0167] Figure 13 A diagram of an apparatus for planning a method for intelligent agent air-ground collaborative tasks according to an embodiment of the present disclosure is shown. Detailed Implementation

[0168] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0169] To simultaneously meet the needs of multi-agent air-ground cooperative task planning, and to ensure both real-time performance and global search capability and stable trajectory control performance, this disclosure provides a method flow 100 for planning intelligent agent air-ground cooperative tasks, such as... Figure 1 As shown, it includes the following steps:

[0170] In step S101, the intelligent agent includes a drone and an unmanned vehicle. The task scenario of the intelligent agent and the two-dimensional kinematic model of the drone and the two-dimensional kinematic model of the unmanned vehicle are constructed.

[0171] Specifically Figure 2 The illustration shows a schematic diagram of an air-ground collaborative task scenario in an embodiment of this disclosure, which includes multiple drones, unmanned vehicles, and task targets distributed in a two-dimensional plane. UGV and UAV represent unmanned vehicles and drones, respectively, and T1, T2, T3, and T4 are tasks. This illustrates a practical application scenario in which heterogeneous intelligent agents collaboratively perform mobile target tracking and area patrol tasks within a unified task space.

[0172] The task scenarios in step S101 include scenarios of various scales and tasks of various types. The scenarios of various scales include small-scale scenarios, medium-scale scenarios and large-scale scenarios. The task types include dynamic moving target tracking and fixed area patrol and surveillance.

[0173] The task set in the task scenario Represented as: , Indicates the first One task;

[0174] In a two-dimensional task space, the task The starting point and the ending point are represented as follows:

[0175] ;

[0176] In the formula, Let be the two-dimensional coordinates at the start of task j; Let task j start with its horizontal coordinate value; The vertical coordinate value at the start of task j; The two-dimensional coordinates at the end of task j; The horizontal coordinate value at the end of task j; This represents the vertical coordinate value at the end of task j;

[0177] Configure constraints with deadlines. And define a binary task assignment matrix. Represented as:

[0178] ;

[0179] In the formula, Indicates the first One intelligent agent; Represents intelligent agents Assigned to perform tasks , This indicates that the task has not been assigned. Task assignment satisfies the following assignment constraints:

[0180] ;

[0181] ;

[0182] In the formula, For a collection of intelligent agents, The assignment constraint is used to configure each task to be executed by an agent, and each agent executes only one task within the same planning period.

[0183] In step S1, the motion state of the agent in the two-dimensional scene is represented by its planar position and heading angle. planar position for:

[0184] ;

[0185] In the formula, Let be the horizontal coordinate value of agent i; Here is the vertical coordinate value of agent i;

[0186] At the mission planning level, it is assumed that the UAV flies at an approximately constant altitude, ignoring the impact of altitude changes on the planar position of the mission point. Therefore, only the UAV's position at the mission point is considered. For translation and heading changes within a plane, the two-dimensional kinematic model of the UAV includes:

[0187] ;

[0188] In the formula, Let x be the velocity of the drone in the horizontal x-axis direction; Let y be the velocity of the UAV along the vertical axis. The angular velocity of the drone's heading; This refers to level flight speed; For heading angle; This is the yaw angle control input; , For a subset of drones, Based on the aforementioned two-dimensional kinematic model of the UAV, integration is performed on a given task assignment and initial state to obtain the feasible trajectory of the UAV on the two-dimensional plane, as well as the corresponding path length, energy consumption, and arrival time.

[0189] The unmanned vehicle moves in the ground plane, and the constraint relationship between velocity and steering is described using the Ackermann kinematic model, resulting in a two-dimensional kinematic model of the unmanned vehicle including:

[0190] ;

[0191] In the formula, Let x be the velocity of the autonomous vehicle in the horizontal x-axis direction; Let be the velocity of the autonomous vehicle in the vertical y-axis direction; Let be the angular velocity of the autonomous vehicle's heading. For the vehicle's heading angle, For vehicle wheelbase, The driving speed of the driverless car, This refers to the front wheel steering angle; For driverless car subsets, The two-dimensional kinematic model of the autonomous vehicle describes the feasible radius of curvature and steering constraints of the autonomous vehicle in a two-dimensional plane, providing a basis for subsequent evaluation of the feasibility of the autonomous vehicle path and control of energy consumption.

[0192] Next, proceed to step S102.

[0193] In step S102, a complete communication information interaction structure and a non-complete communication information interaction structure of the intelligent agent are constructed.

[0194] Specifically, this embodiment considers both complete and incomplete communication modes to characterize the information interaction structure of a multi-agent system. Under the complete communication information interaction structure in step S102, all agents share global information, and the neighborhood set of an agent is represented as:

[0195] ;

[0196] Under the aforementioned incomplete communication information interaction structure, agents can only communicate with other agents in spatial proximity, within a communication radius. The neighborhood set is defined as follows:

[0197] ;

[0198] In the formula, For agents within the neighborhood of agent i under incomplete communication; The position of agent i; Let j be the position of agent j.

[0199] Next, proceed to step S103.

[0200] In step S103, based on the agent's task scenario, the UAV's two-dimensional kinematic model, and the unmanned vehicle's two-dimensional kinematic model, the agent's cost model and comprehensive objective function are constructed in an integral form, and the local comprehensive cost is obtained.

[0201] Specifically, the cost model in step S103 adopts an integral form, including:

[0202] ;

[0203] ;

[0204] In the formula, and These are the speed and steering inputs for the unmanned vehicle, or the yaw control inputs for the drone; These are the weighting coefficients; For intelligent agents The planar position vector at the current moment; For the task The starting position vector is the position where the task begins execution; For the task The endpoint position vector, which is the position reached after the task is completed; For intelligent agents The final stopping position vector; This represents the Euclidean norm.

[0205] The integrated objective function includes:

[0206] ;

[0207] In the formula, The total time to complete the task; Task deadline; weight The importance of adjusting the path, controlling energy consumption, and penalties for overdue payments; Assign matrix to task The Middle The first agent and the second Binary decision variables between tasks; For intelligent agents Execute the task The path cost at any given time represents the distance the agent travels from its current position to the task. The sum of the path length or distance from the starting point to the destination after completing the task; For intelligent agents Execute the task Cost control.

[0208] Next, proceed to step S104.

[0209] In step S104, based on the local comprehensive cost and introducing the task urgency factor, the bidding results of each agent for each task are obtained. Based on the bidding results, the pre-allocation mechanism of the greedy auction algorithm is adopted to traverse the task scenarios under the complete communication information interaction structure and the incomplete communication information interaction structure respectively, and the task pre-allocation matrix and the corresponding task pre-allocation quality index are obtained.

[0210] Specifically, the bidding results in step S104 include:

[0211] ;

[0212] in It is a regular term; For local comprehensive costs; As a task urgency factor, ; For the current moment, It is a constant, and Task deadline The closer they are, the higher the task urgency factor. The larger;

[0213] For the same task Local comprehensive cost The smaller the bid, the better. The larger the urgency factor, the tighter the deadline for different tasks. The larger the size, the higher the priority it will be in the auction;

[0214] A greedy auction algorithm is used for pre-allocation. For the fully communicative information interaction structure, the bidding results of the agents for the corresponding tasks are auctioned across the entire network. For the incompletely communicative information interaction structure, local bidding is conducted in the neighborhood set. After the local bidding ends, each agent broadcasts its current temporary task allocation result and corresponding cost information to its neighbors in multiple communication rounds. Based on the received neighbor information, the local task allocation variables are updated and conflicts are resolved. For example, when the same task is won by multiple agents at the same time or a single agent is assigned multiple tasks, the agent compares the bids or the cost retention cost and selects the one with the better cost until the decisions of the agents in the neighborhood regarding task ownership no longer change, resulting in a consistent task pre-allocation matrix across the entire network.

[0215] Figure 3 A flowchart of the Greedy Auction Algorithm (GAAI) according to an embodiment of this disclosure is shown, as follows: Figure 3 As shown, the bidding process includes:

[0216] Step A1: Initialize the set of unassigned tasks A collection of intelligent agents is available. ;

[0217] Step A2: In all Find the highest bid result:

[0218] ;

[0219] In the formula, For tasks in the unassigned task set; Agents are those in the set of available agents;

[0220] In the context of a fully interactive communication structure, A represents the set of all intelligent agents; in the context of a partially interactive communication structure, A represents the set of intelligent agents participating in the auction within the current communication neighborhood.

[0221] Step A3: Based on the maximum bid result, allocate tasks from the unassigned task set. Agents assigned to the set of available agents ,set up and from the set of available agents and unassigned task set Remove the corresponding agent or task from the list;

[0222] Step A4: Repeat steps A2–A3 until all tasks have been assigned or no agents are available, to obtain the initial assignment scheme;

[0223] Step A5: Based on the pre-allocation results and local comprehensive costs Obtain the task pre-assignment matrix and task pre-assignment matrix The corresponding global objective function value for:

[0224] ;

[0225] In the formula, Pre-allocate matrix for tasks The Middle Line 1 Column elements, It is a binary decision variable, representing the agent. With the task In the pre-allocation phase, the association relationship is such that a value of 1 indicates allocation, and a value of 0 indicates no allocation.

[0226] Step A6: Based on the global objective function value Construct task pre-allocation quality indicators :

[0227] ;

[0228] In the formula, These represent the empirical optimal and empirical worst values ​​in the feasible solution space under a given task scenario, respectively. Based on the empirical optimal value... With worst experience It characterizes the relative merits of the current initial solution.

[0229] Next, proceed to step S105.

[0230] In step S105, the improved whale optimization algorithm is used to optimize the initial allocation scheme according to the task pre-allocation quality index to obtain the optimal task allocation matrix, where the initial allocation scheme is the task pre-allocation matrix.

[0231] Specifically Figure 4 The diagram illustrates a greedy auction algorithm combined with an improved whale optimization algorithm (GAAI-IWOA) according to an embodiment of this disclosure. It shows that GAAI first generates an initial pre-allocation matrix, and then the IWOA module, based on the initial solution, performs a global search and refinement optimization of the task pre-allocation matrix through mechanisms such as encirclement strategy, spiral update, and optimal solution replacement. In step S105, an adaptive nonlinear convergence factor, a Gaussian perturbation local development strategy, and an optimal solution replacement strategy are introduced into the whale algorithm to obtain the improved whale optimization algorithm, which is then based on the task pre-allocation quality index. The optimal task allocation matrix is ​​obtained by globally optimizing the task pre-assignment matrix. Among them, the adaptive nonlinear convergence factor and Gaussian perturbation local exploitation strategy dynamically adjust the balance between exploration and exploitation during the iteration process, thereby improving search diversity and convergence stability, and overcoming the shortcomings of standard WOA in discrete task allocation.

[0232] The adaptive nonlinear convergence factor Based on task pre-allocation quality indicators Adjusting the convergence factor exponent includes the following steps:

[0233] ;

[0234] In the formula, To adjust the parameters; This represents the maximum number of iterations. for To be based on pre-allocated quality indicators; A changing power function; This is the initial value of the exponent;

[0235] The local exploitation strategy for the Gaussian perturbation includes:

[0236] ;

[0237] In the formula, , The current iteration algebra The position vector of the individual with the lowest fitness in the next population is the position corresponding to the global optimal solution in the current iteration; For the first The era The position vector of the nth individual, corresponding to the nth individual Encoded representation of each candidate task allocation scheme; It is a spiral contraction factor; The helical coefficient is... Randomly generated, Controlling noise amplitude Let be a standard Gaussian random variable with a mean of 0 and a variance of 1;

[0238] The optimal solution replacement strategy includes: replacing the worst solution with the current optimal solution after each iteration. ,but ;

[0239] In the formula, This is the worst solution; This is the optimal solution;

[0240] The improved whale optimization algorithm is based on parameters Determine whether to perform a bounding or searching operation, and then update the entire system accordingly.

[0241] ;

[0242] In the formula, In order to be in Uniformly distributed random numbers over an interval For coefficient vectors;

[0243] like Execute the encirclement strategy;

[0244] ;

[0245] like Randomly select individuals search

[0246] ;

[0247] In the formula, This indicates element-wise multiplication. Individuals randomly selected from the current population; For the first Era Intelligent Agent The position vector, and the corresponding encoded representation of the candidate task allocation scheme;

[0248] When combining three improvement strategies to obtain the optimal task allocation matrix, the following are included:

[0249] Step B1: Assign the task pre-matrix Mapped to individual whales according to a preset encoding method. and in the task pre-assignment matrix Several individuals are generated within the neighborhood of the whale, forming the initial population for the improved whale optimization algorithm; each individual... Each corresponds to a candidate task assignment matrix.

[0250] Step B2: For any iterative algebra Individuals in Decoding yields the corresponding current task allocation matrix. And calculate the fitness based on the comprehensive objective function:

[0251] ;

[0252] in, Using the comprehensive objective function defined in step S3, sort the individuals by fitness values ​​from smallest to largest to obtain the current optimal individual. and its corresponding task allocation matrix and the worst individual at present .

[0253] Step B3: In the In the iteration, the given adaptive nonlinear convergence factor is first used. Calculate the parameter vector:

[0254] ;

[0255] In the formula, In order to be in Uniformly distributed random numbers over an interval For coefficient vectors;

[0256] when At this time, corresponding to the encirclement and local development stages: for individuals closest to the current optimal individual individual First calculate Based on the standard bounding update, a local exploitation strategy for the defined Gaussian perturbation is introduced to update the individual positions:

[0257] ;

[0258] In the formula, , It is a spiral contraction factor; The helical coefficient; The number is randomly generated; These are noise amplitude control parameters; Let be a standard Gaussian random variable with a mean of 0 and a variance of 1; by superimposing Gaussian perturbations in the enclosing operation, the search for the neighborhood of the current optimal task assignment matrix is ​​achieved;

[0259] when At this time, corresponding to the global search phase: randomly select individuals. As a search reference, for individuals Update:

[0260] ;

[0261] In the formula, This indicates element-wise multiplication. Individuals randomly selected from the current population are used to expand the global exploration range of the solution space;

[0262] Step B4: Complete for all individuals After the position is updated, the fitness of the new population is recalculated to obtain the best individuals of the new generation. With the worst individual According to the given optimal solution replacement strategy, if Then perform the replacement operation. Ensure that at least one currently optimal task allocation scheme is retained in each generation of the population;

[0263] Step B5: Repeat steps B2 to B4 until the preset maximum number of iterations is reached. Termination conditions include fitness changes falling below a preset threshold.

[0264] At termination, the individual with the lowest fitness is selected from the final population. and will Decoding yields the corresponding optimal task allocation matrix. The optimal task allocation matrix This represents the optimal task allocation result.

[0265] Next, proceed to step S106.

[0266] In step S106, a UAV prediction model and an unmanned vehicle prediction model are constructed, along with the agent's reference output trajectory and reference control sequence. Combined with the optimal task allocation matrix, the final planning result is obtained.

[0267] Specifically Figure 5 The trajectory optimization control flow of an embodiment of this disclosure is illustrated, such as... Figure 5 As shown, the UAV prediction model in step S106 includes:

[0268] A state-space-based discrete-time prediction model is used to model the UAV, and the UAV prediction model includes:

[0269] ;

[0270] In the formula, Index for drones; For drones at discrete time The state vector; This is the control input vector for the UAV. This is the output vector of the UAV; Used to select the UAV's planar position from the state vector as the output; This is the discrete-time system matrix of the UAV, which is based on the UAV's two-dimensional kinematic model after sampling period. Linearization and discretization yield:

[0271] The continuous-time nonlinear model can be written as:

[0272] ;

[0273] In the formula, ;

[0274] Select sampling period The frequency is dynamically determined by the system control frequency and the actuator, and generally ranges from tens to hundreds of milliseconds.

[0275] In the reference trajectory For nonlinear functions Linearization is performed using a first-order Taylor expansion, and zero-order preservation or Euler approximation is used according to the sampling period. Discretize the system and obtain the discrete-time system matrix according to linear system theory:

[0276] ;

[0277] In the formula, , For the selected linearized operating point, It is a third-order identity matrix; , These are the Jacobian matrices for the state and control, respectively. In this embodiment, the method for constructing the UAV prediction model is a common approach for establishing discrete linear prediction models near a reference trajectory for continuous-time nonlinear systems. This embodiment directly uses the theoretical results to construct the UAV prediction model. .

[0278] The autonomous vehicle prediction model includes:

[0279] The autonomous vehicle prediction model is constructed using a state-space-based discrete-time prediction model, including:

[0280] ;

[0281] in, Index for driverless cars; Used to select the position of the unmanned vehicle in the plane as the output from the state vector; For autonomous vehicles at discrete times The state vector; This is the control input vector for the autonomous vehicle; This is the output vector of the autonomous vehicle; This is the discrete-time system matrix of the autonomous vehicle, which is based on the two-dimensional kinematic model of the autonomous vehicle after sampling period. Linearization and discretization yield:

[0282] The continuous-time nonlinear model can be written as:

[0283] ;

[0284] In the formula, , is the vehicle wheelbase;

[0285] In the reference trajectory For nonlinear functions Linearization is performed using a first-order Taylor expansion, and zero-order preservation or Euler approximation is used according to the sampling period. Discretize the system and, according to linear system theory, obtain the discrete-time system matrix:

[0286] ;

[0287] In the formula, For the selected linearized operating point, It is a third-order identity matrix. , These are the Jacobian matrices for the state and control, respectively. It should be noted that in this embodiment, the construction method of the unmanned vehicle prediction model is consistent with that of the unmanned aerial vehicle (UAV). Both employ the classic method of linearizing a continuous-time nonlinear model to first-order linearization near a linearized reference trajectory and then discretizing it according to the sampling period, thereby obtaining the unmanned vehicle prediction model used for model predictive control. .

[0288] The agent's reference output trajectory includes:

[0289] ;

[0290] In the formula, Indicates the current control cycle Starting from the predicted step size The desired planar position reference trajectory point at any given time. To predict the length of the time domain;

[0291] The reference control sequence includes:

[0292] ;

[0293] In the formula, This indicates the prediction step size in order to achieve the reference output trajectory. The desired control input reference value to be applied at any given time; where and Collectively referred to as The reference control sequence represents the desired speed and steering / yaw inputs for achieving the reference output trajectory under ideal conditions, generated by a simple path-following law or smoothing.

[0294] The process of obtaining the path planning result includes:

[0295] In each control cycle Inside, in the current state Using the initial conditions and the aforementioned UAV prediction model or unmanned vehicle prediction model as constraints, a finite-time optimization problem is constructed:

[0296] ;

[0297] When the state constraints are satisfied Control constraints Under the given conditions, the optimal control sequence is obtained. and the first control variable in the optimal control sequence The path planning result of the current control cycle is output and applied to the actual system, and the state at the next moment is updated by the UAV prediction model or the unmanned vehicle prediction model. ;

[0298] The entire task execution process includes Each control cycle, through Repeat steps C1 to C3 within each control cycle to obtain the complete control quantity sequence and path planning results. The specific steps are as follows:

[0299] Step C1: In the control cycle At the beginning, the state is measured or estimated at the current time. As initial conditions, the finite-time optimization problem is constructed and solved to obtain the optimal control sequence in the prediction time domain. ;

[0300] Step C2: Use the first control variable in the optimal control sequence as the actual control input for the current cycle:

[0301] ;

[0302] And Send to the execution layer of the drone or unmanned vehicle, and record simultaneously. and the resulting state As part of the path planning results;

[0303] Step C3: Update the system state using the discrete-time prediction model to obtain the state at the next sampling time:

[0304] ;

[0305] And order Enter the next control cycle and repeat steps C1 to C3 until... The control input sequence is obtained sequentially:

[0306] ;

[0307] The control input sequence is input into the UAV prediction model or the unmanned vehicle prediction model for recursion to obtain the corresponding state sequence. And by the output equation The calculated output trajectory The output trajectory is used as the complete path planning result from the start time of the task to the end time of the task.

[0308] Summarize all intelligent agents The output trajectory is used to obtain the final path planning result of the multi-agent system in the entire task time domain and is used as the final planning result.

[0309] Figure 6 The following is a dynamic diagram illustrating the effect of GAAI-IWOA under fully communicative conditions according to an embodiment of this disclosure, such as... Figure 6As shown in Figures (a) to (f), the red squares represent the tasks being performed, those surrounded by dashed lines are surveillance tasks, and those without dashed lines are tracking tasks; the star shapes represent unmanned vehicles, the triangles represent drones, and the solid lines represent the obtained path plans. Figure (a) shows the initial task planning result of the GAAI-IWOA algorithm, as well as the initial task and agent positions. Figure (b) shows that the agent has traveled a certain distance according to the previous GAAI-IWOA algorithm results, and it can also be seen that the task planning has not changed significantly, reflecting the stability of the algorithm. Figure (c) shows that the agent is about to enter a large turning phase. Figure (d) shows that the agent has begun the final turning phase, and the task is about to be completed. Figure (e) shows the final stopping position of the agent and the final position of the task. Figure (f) is a final effect rendering, showing the actual task planning result of the agent and task from start to finish. It can be seen that the drone, represented by the yellow triangle, was not assigned any tasks throughout the process, so the distance it moved was negligible. The solid circle represents the final position of the tracking task, while the jumbled, wavy line segments represent the movement trajectory throughout the entire tracking task. Under conditions of full communication, each agent can obtain global task and status information in real time, and the planned trajectory is smooth and feasible. Sufficient safe distances are maintained between agents, and dynamic target tracking and area patrol tasks are completed sequentially, intuitively verifying the feasibility of the proposed task allocation and trajectory planning method.

[0310] To verify the effectiveness of the GAAI-IWOA cooperative task allocation and trajectory planning method in the embodiments of this disclosure, several representative scenarios were constructed in a simulation environment. Based on the different agents and task scales, the simulation scenarios were divided into three categories:

[0311] Small-scale scenario: includes 3 drones, 2 unmanned vehicles, and 5 tasks;

[0312] Medium-scale scenario: includes 5 drones, 4 unmanned vehicles, and 10 tasks;

[0313] Large-scale scenario: includes 10 drones, 6 unmanned vehicles, and 20 tasks.

[0314] The tasks are categorized into two main types: tracking of dynamically moving targets and patrolling and monitoring of fixed areas. Different tasks have different time windows and deadlines to reflect their urgency and time constraints. Simulations are conducted under both full communication information interaction and incomplete communication information interaction modes.

[0315] To ensure a fair comparison, the following classic intelligent optimization algorithms were selected as comparison objects: Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), Differential Evolution (DE), Standard Whale Optimization (WOA), and Falcon Optimization (HHO). Each algorithm was run under the same task scenario and constraints, and the parameters were adjusted to the configuration with better convergence performance.

[0316] For each algorithm, the following performance metrics were calculated:

[0317] Total path length: The sum of the path lengths of all agents during the execution of the assigned task;

[0318] Total control cost: the sum of the square integrals of the control input energy or the control quantity;

[0319] Comprehensive objective function value Based on the weighted objective function in step S103, the path cost, control cost, and overdue penalty are uniformly measured.

[0320] Algorithm running time The computation time required to complete one task allocation optimization;

[0321] Task overdue rate: The percentage of tasks not completed within the specified time window out of the total number of tasks;

[0322] Repeat execution rate: The proportion of the same task that is repeatedly executed by multiple agents.

[0323] In full communication mode, all agents can share global task information and state information, and the comparison results are as follows: Figure 7 – Figure 12 As shown, where, Figure 7 The diagrams show a comparison of path cost, control cost, and total cost for different algorithms in a small-scale scenario. Figure (a) shows a comparison of Path values ​​for each algorithm, Figure (b) shows a comparison of Control×40 values ​​for each algorithm, and Figure (c) shows a comparison of S values ​​for each algorithm. It can be seen that the GAAI-IWOA method proposed in this embodiment significantly outperforms GA, PSO, ACO, DE, WOA, and HHO in all three cost metrics, and has the lowest overall objective function value J(A), indicating that it achieves a better trade-off between path length and energy consumption. Figure 8The corresponding algorithm runtime, task duplication rate, and overdue rate are shown. Figure (a) shows a comparison of the runtime of each algorithm, Figure (b) shows a comparison of the duplication rate of each algorithm, and Figure (c) shows a comparison of the error rate of each algorithm. The runtime of GAAI-IWOA proposed in this embodiment is slightly longer than that of the simplest heuristic algorithm, but still far below the upper limit of the real-time control cycle, meeting the requirements of online applications. At the same time, its task overdue rate and duplication rate are both controlled below 5%, significantly lower than the 10% to 20% level of other algorithms.

[0324] Figure 9 The cost comparison of various algorithms in a medium-scale scenario is shown. Figure (a) shows a comparison of the Path values ​​of each algorithm, Figure (b) shows a comparison of the Control×40 values ​​of each algorithm, and Figure (c) shows a comparison of the S values ​​of each algorithm. As the number of tasks and the scale of the agent increase, the path cost and control cost of traditional algorithms increase to varying degrees. The GAAI-IWOA proposed in this embodiment still maintains the lowest total cost, indicating that it still has good scalability at a larger scale. Figure 10 Figure (a) shows a comparison of the running times of each algorithm, Figure (b) shows a comparison of the duplication rates of each algorithm, and Figure (c) shows a comparison of the error rates of each algorithm. In medium-scale scenarios, the GAAI-IWOA proposed in this embodiment can still control the task overdue rate and duplicate execution rate to within approximately 5%, while the performance degradation of other algorithms is more significant, with overdue and duplicate rates mostly exceeding 15%. Furthermore, the running time of the GAAI-IWOA proposed in this embodiment increases linearly and slowly with increasing scale, still within the acceptable real-time range.

[0325] In large-scale scenarios, such as Figure 11 As shown in the figures, Figure (a) shows a comparison of the Path values ​​of each algorithm, Figure (b) shows a comparison of the Control×40 values ​​of each algorithm, and Figure (c) shows a comparison of the S values ​​of each algorithm. The advantages of the proposed GAAI-IWOA in this embodiment of the present disclosure in terms of path cost and control cost are further highlighted. The total cost is significantly reduced compared to classic algorithms such as GA and PSO, and also shows significant improvement compared to standard WOA and HHO. This indicates that the collaborative framework of introducing greedy auction initialization and improving WOA can effectively suppress combinatorial explosion and premature convergence problems in large-scale task allocation. Figure 12Figure (a) shows a comparison of the running time of each algorithm, Figure (b) shows a comparison of the repetition rate of each algorithm, and Figure (c) shows a comparison of the error rate of each algorithm. Although large-scale scenarios place higher demands on computational complexity, the running time of the proposed GAAI-IWOA in this embodiment is still controlled within an acceptable range, and it maintains a low task overdue rate and repetition rate, demonstrating that the proposed method has good real-time performance and robustness under fully communicative conditions.

[0326] comprehensive Figure 7 – Figure 12 As can be seen, under fully communicative conditions, the GAAI-IWOA method proposed in this embodiment achieves the lowest overall cost and significantly better task execution quality than the comparative algorithms in all three scale scenarios.

[0327] In incomplete communication mode, communication is only allowed within the communication radius. Local information exchange takes place within the domain, and task cost and state estimation need to be gradually propagated and converged within the neighborhood through a consensus algorithm. Although the communication radius is limited and information exchange is incomplete, the communication topology and consensus update mechanism in this embodiment can still achieve optimized performance under near-complete communication conditions, maintaining low cost and high task completion quality in various scale scenarios.

[0328] A second embodiment of the present invention provides a system for planning intelligent agent air-ground cooperative tasks, comprising:

[0329] The initialization unit is configured to: construct the task scenario of the intelligent agent, including drones and unmanned vehicles, and the two-dimensional kinematic model of the drones and the two-dimensional kinematic model of the unmanned vehicles;

[0330] The communication unit is configured to: construct a complete communication information interaction structure and a non-complete communication information interaction structure for the intelligent agent;

[0331] The cost unit is configured as follows: based on the agent's task scenario, the UAV's two-dimensional kinematic model, and the unmanned vehicle's two-dimensional kinematic model, the cost model and comprehensive objective function of the agent are constructed in an integral form, and the local comprehensive cost is obtained.

[0332] The pre-allocation unit is configured to: obtain the bidding results of each agent for each task based on the local comprehensive cost and by introducing a task urgency factor; and, based on the bidding results, adopt a pre-allocation mechanism using a greedy auction algorithm to traverse the task scenarios under both fully communicative information interaction structures and incompletely communicative information interaction structures to obtain a task pre-allocation matrix and corresponding task pre-allocation quality indicators.

[0333] The optimal allocation unit is configured to: use an improved whale optimization algorithm to optimize the initial allocation scheme according to the task pre-allocation quality index to obtain the optimal task allocation matrix, wherein the initial allocation scheme is the task pre-allocation matrix;

[0334] The output unit is configured to: construct a UAV prediction model and an unmanned vehicle prediction model, as well as an agent reference output trajectory and a reference control sequence, and combine them with the optimal task allocation matrix to obtain the final planning result.

[0335] A third embodiment of the present invention also provides an electronic device, the electronic device comprising:

[0336] At least one processor; and,

[0337] The memory is communicatively connected to the at least one processor; wherein,

[0338] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method for planning intelligent agent air-ground collaborative tasks in any of the foregoing embodiments.

[0339] The fourth embodiment of the present invention also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the method for planning intelligent agent air-ground cooperative tasks as described in any of the foregoing embodiments.

[0340] The fifth embodiment of the present invention also provides a computer program product, which includes a computing program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions that, when executed by a computer, cause the computer to perform the method for planning intelligent agent air-ground cooperative tasks according to any of the foregoing embodiments.

[0341] Figure 13 The illustration shows a method or device 1000 implementing an embodiment of the present invention. In some embodiments, more or fewer devices may be included than illustrated. In some embodiments, it may be implemented using a single or multiple devices. In some embodiments, it may be implemented using cloud-based or distributed devices.

[0342] like Figure 13As shown, device 1000 includes a processor 1001, which can perform various appropriate operations and processes based on programs and / or data stored in read-only memory (ROM) 1002 or programs and / or data loaded from storage portion 1008 into random access memory (RAM) 1003. Processor 1001 may be a multi-core processor or may contain multiple processors. In some embodiments, processor 1001 may include a general-purpose main processor and one or more special coprocessors, such as a central processing unit (CPU), graphics processing unit (GPU), neural network processor (NPU), digital signal processor (DSP), etc. Various programs and data required for the operation of device 1000 are also stored in RAM 1003. Processor 1001, ROM 1002, and RAM 1003 are interconnected via bus 1004. Input / output (I / O) interface 1005 is also connected to bus 1004.

[0343] The processor and memory work together to execute a program stored in the memory, which, when executed by a computer, can implement the methods, steps, or functions described in the various embodiments.

[0344] The following components are connected to I / O interface 1005: an input section 1006 including a keyboard, mouse, touchscreen, etc.; an output section 1007 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN card, modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to I / O interface 1005 as needed. A removable medium 1011, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 1010 as needed so that computer programs read from it can be installed into storage section 1008 as needed. Figure 13 The diagram only shows a portion of the components and does not imply that the device 1000 only includes... Figure 13 The components shown.

[0345] The systems, apparatus, modules, or units described in the embodiments can be implemented by a computer or its associated components. The computer may be, for example, a mobile terminal, smartphone, personal computer, laptop computer, in-vehicle human-machine interface device, personal digital assistant, media player, navigation device, game console, tablet computer, wearable device, smart TV, Internet of Things system, smart home, industrial computer, server, or a combination thereof.

[0346] Although not shown, in this embodiment of the invention, a computer-readable storage medium is provided having a computer program / instruction stored thereon, which, when executed by a processor, implements the method for planning intelligent agent air-ground cooperative tasks as described in the embodiment.

[0347] Storage media in embodiments of the present invention include articles that are permanent and non-permanent, removable and non-removable, capable of storing information by any method or technology. Examples of storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0348] Although not shown, embodiments of the present invention also provide a computer program product, including: a computer program / instructions that, when executed by a processor, implement the method for planning intelligent agent air-ground collaborative tasks as described in the embodiments.

[0349] The methods, programs, systems, apparatuses, etc., in embodiments of the present invention can be executed or implemented in one or more networked computers, or practiced in a distributed computing environment. In the embodiments of this specification, in these distributed computing environments, tasks can be performed by remote processing devices connected via a communication network.

Claims

1. A method for planning intelligent agent air-ground cooperative tasks, characterized in that, Includes the following steps: Step S1: The intelligent agent includes drones and unmanned vehicles. Construct the task scenario of the intelligent agent and the two-dimensional kinematic model of the drone and the two-dimensional kinematic model of the unmanned vehicle. Step S2: Construct the complete communication information interaction structure and the incomplete communication information interaction structure of the intelligent agent; Step S3: Based on the intelligent agent task scenario, the two-dimensional kinematic model of the UAV, and the two-dimensional kinematic model of the unmanned vehicle, construct the cost model and comprehensive objective function of the intelligent agent in an integral form, and obtain the local comprehensive cost; Step S4: Based on the local comprehensive cost and introducing the task urgency factor, obtain the bidding results of each agent for each task. Based on the bidding results, adopt the pre-allocation mechanism of the greedy auction algorithm, traverse the task scenarios under the complete communication information interaction structure and the incomplete communication information interaction structure respectively, and obtain the task pre-allocation matrix and the corresponding task pre-allocation quality index. Step S5: The improved whale optimization algorithm is used to optimize the initial allocation scheme according to the task pre-allocation quality index to obtain the optimal task allocation matrix. The initial allocation scheme is the task pre-allocation matrix. Step S6: Construct the UAV prediction model and the unmanned vehicle prediction model, as well as the intelligent agent's reference output trajectory and reference control sequence, and combine them with the optimal task allocation matrix to obtain the final planning result.

2. The method for planning intelligent agent air-ground cooperative tasks according to claim 1, characterized in that, The task scenarios in step S1 include scenarios of various scales and tasks of various types. The scenarios of various scales include small-scale scenarios, medium-scale scenarios and large-scale scenarios. The task types include dynamic moving target tracking and fixed area patrol and surveillance. The task set in the task scenario Represented as: , Indicates the first One task; In a two-dimensional task space, the task The starting point and the ending point are represented as follows: ； In the formula, Let be the two-dimensional coordinates at the start of task j; Let task j start with its horizontal coordinate value; The vertical coordinate value at the start of task j; The two-dimensional coordinates at the end of task j; The horizontal coordinate value at the end of task j; This represents the vertical coordinate value at the end of task j; Configure constraints with deadlines. And define a binary task assignment matrix. Represented as: ； In the formula, Indicates the first One intelligent agent; Represents intelligent agents Assigned to perform tasks , This indicates that the task has not been assigned. Task assignment satisfies the following assignment constraints: ； ； In the formula, For a collection of intelligent agents, ; The assignment constraint is used to configure each task to be executed by an agent, and each agent executes only one task within the same planning period; In step S1, the motion state of the agent in the two-dimensional scene is represented by its planar position and heading angle. planar position for: ； In the formula, Let be the horizontal coordinate value of agent i; Here is the vertical coordinate value of agent i; The drone flies at a constant altitude in the air, considering only the drone's position... For translation and heading changes within a plane, the two-dimensional kinematic model of the UAV includes: ； In the formula, Let x be the velocity of the drone in the horizontal x-axis direction; Let y be the velocity of the UAV along the vertical axis. The angular velocity of the drone's heading; This refers to level flight speed; For heading angle; This is the yaw angle control input; , For a subset of drones, ; The unmanned vehicle moves in the ground plane, and the constraint relationship between velocity and steering is described using the Ackermann kinematic model, resulting in a two-dimensional kinematic model of the unmanned vehicle including: ； In the formula, Let x be the velocity of the autonomous vehicle in the horizontal x-axis direction; Let be the velocity of the autonomous vehicle in the vertical y-axis direction; Let be the angular velocity of the autonomous vehicle's heading. For the vehicle's heading angle, For vehicle wheelbase, The driving speed of the driverless car, This refers to the front wheel steering angle; For driverless car subsets, .

3. The method for planning intelligent agent air-ground cooperative tasks according to claim 1, characterized in that, Under the fully communicative information interaction structure in step S2, all agents share global information, and the neighborhood set of an agent is represented as follows: ； Under the aforementioned incomplete communication information interaction structure, agents can only communicate with other agents in spatial proximity, within a communication radius. The neighborhood set is defined as follows: ； In the formula, For agents within the neighborhood of agent i under incomplete communication; The position of agent i; Let j be the position of agent j.

4. The method for planning intelligent agent air-ground cooperative tasks according to claim 1, characterized in that, The cost model in step S3 includes: ； ； In the formula, and These are the speed and steering inputs for the unmanned vehicle, or the yaw control inputs for the drone; These are the weighting coefficients; For intelligent agents The planar position vector at the current moment; For the task The starting position vector; For the task The endpoint position vector; For intelligent agents The final stopping position vector; The integrated objective function includes: ； In the formula, The total time to complete the task; Task deadline; weight The importance of adjusting the path, controlling energy consumption, and penalties for overdue payments; Assign matrix to task The Middle The first agent and the second Binary decision variables between tasks; For intelligent agents Execute the task The path cost at any given time represents the distance the agent travels from its current position to the task. The sum of the path length or distance from the starting point to the destination after completing the task; For intelligent agents Execute the task Cost control.

5. The method for planning intelligent agent air-ground cooperative tasks according to claim 1, characterized in that, The bidding results in step S4 include: ； in It is a regular term; For local comprehensive costs; As a task urgency factor, ; For the current moment, It is a constant, and Task deadline The closer they are, the higher the task urgency factor. The larger; For the same task Local comprehensive cost The smaller the bid, the better. The larger the urgency factor, the tighter the deadline for different tasks. The larger the size, the higher the priority it will be in the auction; A greedy auction algorithm is used for pre-allocation. For the fully communicative information interaction structure, the bidding results of the agents for the corresponding tasks are auctioned across the entire network. For the incompletely communicative information interaction structure, local bidding is conducted in the neighborhood set. After the local bidding ends, each agent broadcasts its current temporary task allocation result and corresponding cost information to its neighbors in multiple communication rounds. Based on the received neighbor information, the agent updates the local task allocation variables and resolves conflicts until the decisions of each agent in the neighborhood regarding task ownership no longer change, resulting in a consistent task pre-allocation matrix across the entire network. The bidding process includes: Step A1: Initialize the set of unassigned tasks A collection of intelligent agents is available. ; Step A2: In all Find the highest bid result: ； In the formula, For tasks in the unassigned task set; Agents are those in the set of available agents; In the context of a fully interactive communication structure, A represents the set of all intelligent agents; in the context of a partially interactive communication structure, A represents the set of intelligent agents participating in the auction within the current communication neighborhood. Step A3: Based on the maximum bid result, allocate tasks from the unassigned task set. Agents assigned to the set of available agents ,set up and from the set of available agents and unassigned task set Remove the corresponding agent or task from the list; Step A4: Repeat steps A2–A3 until all tasks have been assigned or no agents are available, to obtain the initial assignment scheme; Step A5: Based on the pre-allocation results and local comprehensive costs Obtain the task pre-assignment matrix and task pre-assignment matrix The corresponding global objective function value for: ； In the formula, Pre-allocate matrix for tasks The Middle Line 1 Column elements, It is a binary decision variable, representing the agent. With the task In the pre-allocation phase, the association relationship is such that a value of 1 indicates allocation, and a value of 0 indicates no allocation. Step A6: Based on the global objective function value Construct task pre-allocation quality indicators : ； In the formula, These represent the empirical optimal and empirical worst values ​​in the feasible solution space under a given task scenario, respectively. Based on the empirical optimal value... With worst experience It characterizes the relative merits of the current initial solution.

6. The method for planning intelligent agent air-ground cooperative tasks according to claim 1, characterized in that, In step S5, an adaptive nonlinear convergence factor, a Gaussian perturbation local exploitation strategy, and an optimal solution replacement strategy are introduced into the whale algorithm to obtain an improved whale optimization algorithm, which is then based on task pre-allocation quality indicators. The optimal task allocation matrix is ​​obtained by globally optimizing the task pre-assignment matrix. ; The adaptive nonlinear convergence factor Based on task pre-allocation quality indicators Adjusting the convergence factor exponent includes the following steps: ； In the formula, To adjust the parameters; This represents the maximum number of iterations. for To be based on pre-allocated quality indicators; A changing power function; This is the initial value of the exponent; The local exploitation strategy for the Gaussian perturbation includes: ； In the formula, , The current iteration algebra The position vector of the individual with the lowest fitness in the next population is used as the position corresponding to the global optimal solution in the current iteration; For the first The era The position vector of the nth individual, corresponding to the nth individual Encoded representation of each candidate task allocation scheme; It is a spiral contraction factor; The helical coefficient is... Randomly generated, Controlling noise amplitude Let be a standard Gaussian random variable with a mean of 0 and a variance of 1; The optimal solution replacement strategy includes: replacing the worst solution with the current optimal solution after each iteration. ,but ; In the formula, This is the worst solution; This is the optimal solution; The improved whale optimization algorithm is based on parameters Determine whether to perform a bounding or searching operation, and then update the entire system accordingly. ； In the formula, In order to be in Uniformly distributed random numbers over an interval For coefficient vectors; like Execute the encirclement strategy; ； like Randomly select individuals search ； In the formula, This indicates element-wise multiplication. Individuals randomly selected from the current population; For the first Era Intelligent Agent The position vector, and the corresponding encoded representation of the candidate task allocation scheme; When combining three improvement strategies to obtain the optimal task allocation matrix, the following are included: Step B1: Assign the task pre-matrix Mapped to individual whales according to a preset encoding method. and in the task pre-assignment matrix Several individuals are generated within the neighborhood of the whale, forming the initial population for the improved whale optimization algorithm; each individual... Each corresponds to a candidate task assignment matrix; Step B2: For any iterative algebra Individuals in Decoding yields the corresponding current task allocation matrix. And calculate the fitness based on the comprehensive objective function: ； in, Using the comprehensive objective function defined in step S3, sort the individuals by fitness values ​​from smallest to largest to obtain the current optimal individual. and its corresponding task allocation matrix and the worst individual at present ; Step B3: In the In the iteration, the given adaptive nonlinear convergence factor is first used. Calculate the parameter vector: ； In the formula, In order to be in Uniformly distributed random numbers over an interval For coefficient vectors; when At this time, corresponding to the encirclement and local development stages: for individuals closest to the current optimal individual individual First calculate Based on the standard bounding update, a local exploitation strategy for the defined Gaussian perturbation is introduced to update the individual positions: ； In the formula, , It is a spiral contraction factor; The helical coefficient; The number is randomly generated; These are noise amplitude control parameters; Let be a standard Gaussian random variable with a mean of 0 and a variance of 1; by superimposing Gaussian perturbations in the enclosing operation, the search for the neighborhood of the current optimal task assignment matrix is ​​achieved; when At this time, corresponding to the global search phase: randomly select individuals. As a search reference, for individuals Update: ； In the formula, This indicates element-wise multiplication. Individuals randomly selected from the current population are used to expand the global exploration range of the solution space; Step B4: Complete for all individuals After the position is updated, the fitness of the new population is recalculated to obtain the best individuals of the new generation. With the worst individual According to the given optimal solution replacement strategy, if Then perform the replacement operation. Ensure that at least one currently optimal task allocation scheme is retained in each generation of the population; Step B5: Repeat steps B2 to B4 until the preset maximum number of iterations is reached. Termination conditions include fitness changes falling below a preset threshold. At termination, the individual with the lowest fitness is selected from the final population. and will Decoding yields the corresponding optimal task allocation matrix. The optimal task allocation matrix This represents the optimal task allocation result.

7. The method for planning intelligent agent air-ground cooperative tasks according to claim 1, characterized in that, In step S6, the UAV prediction model includes: A state-space-based discrete-time prediction model is used to model the UAV, and the UAV prediction model includes: ； In the formula, Index for drones; For drones at discrete time The state vector; This is the control input vector for the UAV. This is the output vector of the UAV; Used to select the UAV's planar position from the state vector as the output; This is the discrete-time system matrix of the UAV, which is based on the UAV's two-dimensional kinematic model after sampling period. Linearization and discretization yield: The continuous-time nonlinear model can be written as: ； In the formula, ; In the reference trajectory For nonlinear functions Linearization is performed using a first-order Taylor expansion, and zero-order preservation or Euler approximation is used according to the sampling period. Discretize the system and obtain the discrete-time system matrix according to linear system theory: ； In the formula, , For the selected linearized operating point, It is a third-order identity matrix; , These are the Jacobian matrices for the state and control, respectively; The autonomous vehicle prediction model includes: The autonomous vehicle prediction model is constructed using a state-space-based discrete-time prediction model, including: ； in, Index for driverless cars; Used to select the position of the unmanned vehicle in the plane as the output from the state vector; For autonomous vehicles at discrete times The state vector; This is the control input vector for the autonomous vehicle; This is the output vector of the autonomous vehicle; This is the discrete-time system matrix of the autonomous vehicle, which is based on the two-dimensional kinematic model of the autonomous vehicle after sampling period. Linearization and discretization yield: The continuous-time nonlinear model can be written as: ； In the formula, L is the vehicle wheelbase; In the reference trajectory For nonlinear functions Linearization is performed using a first-order Taylor expansion, and zero-order preservation or Euler approximation is used according to the sampling period. Discretize the system and, according to linear system theory, obtain the discrete-time system matrix: ； In the formula, For the selected linearized operating point, It is a third-order identity matrix. , These are the Jacobian matrices for the state and control, respectively; The agent's reference output trajectory includes: ； In the formula, Indicates the current control cycle Starting from the predicted step size The desired planar position reference trajectory point at any given time. To predict the length of the time domain; The reference control sequence includes: ； In the formula, This indicates the prediction step size in order to achieve the reference output trajectory. The desired control input reference value to be applied at any given time; where and Collectively referred to as , and Collectively referred to as ; The process of obtaining the path planning result includes: In each control cycle Inside, in the current state Using the initial conditions and the aforementioned UAV prediction model or unmanned vehicle prediction model as constraints, a finite-time optimization problem is constructed: ； When the state constraints are satisfied Control constraints Under the given conditions, the optimal control sequence is obtained. and the first control variable in the optimal control sequence The path planning result of the current control cycle is output and applied to the actual system, and the state at the next moment is updated by the UAV prediction model or the unmanned vehicle prediction model. ; The entire task execution process includes Each control cycle, through Repeat steps C1 to C3 within each control cycle to obtain the complete control quantity sequence and path planning results. The specific steps are as follows: Step C1: In the control cycle At the beginning, the state is measured or estimated at the current time. As initial conditions, the finite-time optimization problem is constructed and solved to obtain the optimal control sequence in the prediction time domain. ; Step C2: Use the first control variable in the optimal control sequence as the actual control input for the current cycle: ； And Send to the execution layer of the drone or unmanned vehicle, and record simultaneously. and the resulting state As part of the path planning results; Step C3: Update the system state using the discrete-time prediction model to obtain the state at the next sampling time: ； And order Enter the next control cycle and repeat steps C1 to C3 until... The control input sequence is obtained sequentially: ； The control input sequence is input into the UAV prediction model or the unmanned vehicle prediction model for recursion to obtain the corresponding state sequence. And by the output equation The calculated output trajectory The output trajectory is used as the complete path planning result from the start time of the task to the end time of the task. Summarize all intelligent agents The output trajectory is used to obtain the final path planning result of the multi-agent system in the entire task time domain and is used as the final planning result.

8. A system for planning intelligent agent air-ground collaborative tasks, characterized in that, The method for planning intelligent agent air-ground cooperative tasks according to any one of claims 1-7 includes: The initialization unit is configured to: construct the task scenario of the intelligent agent, including drones and unmanned vehicles, and the two-dimensional kinematic model of the drones and the two-dimensional kinematic model of the unmanned vehicles; The communication unit is configured to: construct a complete communication information interaction structure and a non-complete communication information interaction structure for the intelligent agent; The cost unit is configured as follows: based on the agent's task scenario, the UAV's two-dimensional kinematic model, and the unmanned vehicle's two-dimensional kinematic model, the cost model and comprehensive objective function of the agent are constructed in an integral form, and the local comprehensive cost is obtained. The pre-allocation unit is configured to: obtain the bidding results of each agent for each task based on the local comprehensive cost and by introducing a task urgency factor; and, based on the bidding results, adopt a pre-allocation mechanism using a greedy auction algorithm to traverse the task scenarios under both fully communicative information interaction structures and incompletely communicative information interaction structures to obtain a task pre-allocation matrix and corresponding task pre-allocation quality indicators. The optimal allocation unit is configured to: use an improved whale optimization algorithm to optimize the initial allocation scheme according to the task pre-allocation quality index to obtain the optimal task allocation matrix, wherein the initial allocation scheme is the task pre-allocation matrix; The output unit is configured to: construct a UAV prediction model and an unmanned vehicle prediction model, as well as an agent reference output trajectory and a reference control sequence, and combine them with the optimal task allocation matrix to obtain the final planning result.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and, The memory is communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method for planning intelligent agent air-ground collaborative tasks as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions for causing the computer to perform the method for planning an intelligent agent's air-ground cooperative task as described in any one of claims 1 to 7.

Images