Two-stage evolutionary multi-robot task allocation method in nuclear accident rescue and electronic equipment

Abstract

The invention relates to the technical field of multi-robot control systems, in particular to a two-stage evolutionary multi-robot task allocation method and electronic equipment in nuclear accident rescue, and the method comprises the following steps: S1, initializing multi-robot and task information; s2, establishing an optimization problem model; s3, multi-robot task allocation; through the two-stage evolutionary algorithm, working time balance distribution between the robots can be comprehensively considered, the maximum radiation accumulation between all the robots and the overall task waiting cost are minimized, and the tasks are distributed to the robots according to the criterion of the minimum comprehensive cost; according to the two-stage evolutionary algorithm used in the method, the population can be quickly converged to the Pareto frontier in the first stage, and the diversity of the solution set is improved in the later stage, so that the algorithm operation time is shortened, and the search efficiency of the globally optimal solution is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-robot control systems, and specifically to a multi-robot task allocation method and an electronic device for nuclear accident rescue with two-stage evolution. Background Art

[0002] Nuclear accident rescue is a highly complex and dangerous task scenario, involving the leakage and diffusion of radioactive substances; traditional human rescue poses great safety risks in such scenarios, while multi-robot systems, with their high efficiency and safety in dangerous environments, have become an ideal choice for nuclear accident rescue; multi-robot systems can not only perform tasks in high-radiation environments, but also improve task completion efficiency through collaborative work; however, how to efficiently allocate tasks so that robots can complete tasks with minimal radiation exposure is a challenging research topic.

[0003] In complex nuclear accident rescue scenarios, the diversity and urgency of tasks further increase the difficulty of task allocation; robots need to perform various tasks in high-radiation areas, such as environmental monitoring, searching for trapped people, handling hazardous substances, etc. These tasks not only have different priorities, but are also subject to time constraints and radiation exposure limits; therefore, a method capable of achieving efficient task allocation under multi-objective and multi-constraint conditions is needed; as a multi-objective optimization method, the two-stage evolutionary algorithm has good global search ability and convergence performance, and can find approximate Pareto optimal solutions in complex optimization problems, which is an effective method to solve the task allocation problem in nuclear accident rescue. Summary of the Invention

[0004] The purpose of the present invention is to provide a multi-robot task allocation method and an electronic device for nuclear accident rescue with two-stage evolution, which can optimize the task allocation of multi-robot systems in complex and dangerous environments.

[0005] Based on the above purpose, the present invention adopts the following technical solutions:

[0006] A multi-robot task allocation method for nuclear accident rescue with two-stage evolution, comprising the following steps:

[0007] S1. Initialize multi-robot and task information: Set the number of robots, the number of tasks, and the number of radiation sources, set information such as the number and execution efficiency of different types of robots and tasks, and obtain the initial positions of robots and task points, as well as the positions of terrain entrances and exits, etc.; Set the scenario instance information according to the task allocation scenario at the nuclear accident rescue site, and specify the problem values;

[0008] S2. Establishment of optimization problem model: Considering various cost requirements in the nuclear accident rescue scenario comprehensively, define the optimization objective, design the encoding of the solution, and represent the solution of the assignment of each task to the robot in an encoded form;

[0009] S3. Multi-robot task assignment: Use a two-stage evolutionary algorithm to optimize the task assignment scheme around the optimization objective in step S2 to obtain a set of solutions for the optimal task assignment scheme.

[0010] Preferably, the scenario instance information in the initialization of multi-robot and task information in step S1 includes:

[0011] (1) RN: The number of robots;

[0012] (2) TN: The number and location of tasks in the area;

[0013] (3) RIN: The number and location of radiation sources;

[0014] (4) KR: The robot type vector, where each element represents the number of robots of the corresponding type;

[0015] (5) KT: The task type vector, where each element represents the number of tasks of the corresponding type;

[0016] (6) E: The efficiency matrix of different robots performing different tasks.

[0017] Preferably, the specific process of defining the optimization objective in the establishment of the optimization problem model in step S2 includes:

[0018] Design the optimization objective to include the robot task execution time, the cumulative radiation received, and the waiting cost, and calculate the optimization objective; the objective function of the problem is expressed as:

[0019]

[0020] Among them, f 1 The goal is to minimize the longest execution time of all robots, f 2 The goal is to minimize the maximum radiation accumulation among all robots, f 3 The goal is to minimize the overall task waiting cost; L k Represents the robot task execution duration, R k Represents the cumulative radiation received by the robot, S k Represents the robot waiting cost, and I represents the set of all robots in the system.

[0021] Preferably, the specific process of calculating the optimization objective includes:

[0022] Calculation of the robot task execution duration:

[0023] Given the task execution order \(P\) of robot \(k\) k , the execution time \(L\) of the robot k is expressed as:

[0024]

[0025] where, represents the task set, is a decision variable, determined by the task execution order \(P\) k ; when robot \(k\) needs to consecutively execute task \(i\) and task \(j\), is 1, otherwise it is 0; \(d\) ij represents the distance between task points, and represent the distances of the task points from the entrance and the exit, represents the efficiency of robot \(k\) in executing task \(i\);

[0026] This formula represents the commuting time spent by the robot entering the scene from the entrance, moving according to the task sequence, and exiting the scene from the exit after the tasks are completed, plus the total time spent by the robot in executing the tasks;

[0027] Calculation of the cumulative radiation received by the robot:

[0028] Given the average radiation intensity \(r\) on each path ij , calculate the radiation accumulated by the robot after passing through this path:

[0029]

[0030] This model describes the cumulative radiation received by the robot when passing through a given path when the positions of the radiation sources are known; given that there are \(N\) s radiation sources in the scene and their positions are known, calculate the average radiation intensity \(r\) ij :

[0031]

[0032] where, and respectively represent the radiation intensities formed by radiation source \(k\) at the starting point and the ending point of the path; the formula for the variation of radiation from a radiation source with distance usually follows the radiation attenuation law, and the variation of the radiation intensity of a ray with distance conforms to the exponential attenuation relationship, expressed as:

[0033] \(I(d)=I\) 0 \(\cdot\exp(-\mu\cdot d)\)

[0034] where \(I(d)\) represents the radiation intensity at a distance \(d\), \(I\) 0is the radiation intensity at the zero distance (i.e., at the radiation source), and μ is the attenuation coefficient of gamma rays;

[0035] Calculation of the robot waiting cost:

[0036] Given the emergency coefficient of each task executed by robot k Then the waiting cost of robot k can be expressed as:

[0037]

[0038] where ts i is the start execution time of task i, obtained by calculating the total time of the tasks executed before task i; this total time includes the sum of the task execution time and the transmission time required to reach the task end point:

[0039]

[0040] where, P k is the task execution sequence of robot k, Robot k executes task efficiency, represents task and task distance between.

[0041] Preferably, in step S2 model establishment, the encoding of the solution adopts a coding method based on the task assignment matrix to accurately represent the assignment relationship between robots and tasks:

[0042] The solution is represented by two components, including a task sequence vector and an entry and exit matrix;

[0043] The task sequence vector includes two basic pieces of information: which tasks each robot executes and the execution order of these tasks;

[0044] The single chromosome technique is used to represent the task sequence vector, and the length of the vector is N t +N r -1, where N t is the number of tasks, and N r is the number of robots.

[0045] Preferably, the specific process of step S3 multi-robot task assignment includes:

[0046] S31. Initialize the population: Generate an initial population so that each individual in the population represents a possible task assignment scheme, and set the population size and the loop termination condition;

[0047] S32. Generate offspring solutions: Calculate the stage parameters; for the initial population in step S31, obtain the offspring solutions using different offspring generation methods according to different stages.

[0048] S33. Environmental selection: Combine the parents and offspring, calculate the individual fitness according to the stage parameters obtained in step S32, and screen the individuals using different screening methods according to the individual fitness.

[0049] S34. Loop output: Check whether the loop termination condition is satisfied. If it is satisfied, output the final population set solution; otherwise, return to step S32.

[0050] Preferably, the loop termination condition in the initial population of step S31 is:

[0051] Calculate the function evaluation times FE = N in the loop, where N is the population size; when the function evaluation times reach the maximum FE max the loop terminates.

[0052] Preferably, the specific process of obtaining the offspring solutions in step S32 includes:

[0053] Calculate the stage parameter I representing the current loop stage stage :

[0054]

[0055] where r c is the parameter controlling the two - level ratio; FE represents the function evaluation times executed by the current algorithm; FE max represents the maximum allowed number of function evaluations;

[0056] According to the different stage parameters, use different offspring generation methods in different stages:

[0057] When I stage = 1:

[0058] The generation of offspring follows the principles of a typical evolutionary algorithm. First, use the binary tournament selection method to obtain the hybridization pool, and then generate offspring based on genetic operators; in the binary tournament selection, the fitness of an individual is determined by its rank in the non - dominated sorting and crowding degree.

[0059] Use the partially - matched crossover PMX method to generate offspring, where the entry and exit matrices of two child nodes are inherited from the parent nodes.

[0060] The mutation operation is set to select two different elements from the task sequence vector for exchange, and the probability of randomly regenerating the entry and exit matrices is 0.2.

[0061] When I stage = 2:

[0062] Perform a local search on the solution with the largest crowding distance within the population, exclude duplicate solutions, and then use the neighbors obtained from this search as the newly generated offspring. After completing the local search, add the solutions to the exploration set; during this stage, use the search set E to track the solutions that have undergone local search.

[0063] The local search uses the 2-opt method to quickly obtain improved solutions; only the task sequence vector is operated on during this stage, while keeping the entrance and exit matrices unchanged.

[0064] Preferably, the specific process of environmental selection in step S33 includes:

[0065] According to the different stage parameters, the environmental selection process is divided into two parts:

[0066] When I stage = 1:

[0067] First, normalize the objective values of the solution set, and then perform non-dominated sorting on the population. Incorporate each rank into the next-generation population in ascending order, starting from the lowest rank until adding the last rank exceeds the population size constraint N;

[0068] Subsequently, according to the fitness of each solution in the last rank, remove the solution with the largest fitness one by one until the population size is reduced to N; during this process, the fitness needs to be recalculated each time a solution is excluded;

[0069] Calculate the fitness of the solution according to its non-dominated rank and crowding distance:

[0070]

[0071] Among them, R represents the non-dominated sorting rank, and C represents the crowding distance; the lower the fitness value, the higher the quality of the solution; the crowding distance of the solution is calculated using the displacement-based density estimation SDE method. When calculating the crowding distance of the current solution, transfer other solutions to the boundary of the dominant region of the current solution; the calculation of the transfer point is shown in the following formula:

[0072]

[0073] Among them, q' is the target value point corresponding to the transfer point, and p is the target point of the solution for which the crowding distance is calculated;

[0074] When I stage = 2:

[0075] The algorithm cancels the population size limit and deletes the dominant solutions and duplicate solutions in the population.

[0076] An electronic device includes a memory and a processor, and a computer program is stored on the memory. It is characterized in that: when the processor executes the computer program, any step in the two-stage evolutionary algorithm for multi-robot task allocation in nuclear accident rescue as described above is implemented.

[0077] The beneficial effects of the present invention are as follows:

[0078] The present invention adopts a two-stage evolutionary algorithm for multi-robot task allocation, takes into account the special requirements in nuclear accident rescue, combines multi-objective optimization problems such as task execution time, radiation accumulation, and waiting cost, can comprehensively consider the balanced distribution of working time among each robot, minimize the maximum radiation accumulation among all robots and the overall task waiting cost, and allocate tasks to all robots according to the criterion of the minimum comprehensive cost, so as to maximize the task execution efficiency of nuclear accident rescue and minimize the radiation impact on robots.

[0079] The two-stage evolutionary algorithm adopted by the present invention can make the population quickly converge to the Pareto front in the first stage and focus on improving the diversity of the solution set in the second stage, thereby reducing the running time of the algorithm, effectively improving the search efficiency of the global optimal solution, significantly improving the efficiency and accuracy of task allocation, and ensuring that robots can efficiently complete tasks on the premise of minimizing radiation exposure. Description of the Drawings

[0080] Figure 1 It is the flowchart of the two-stage evolutionary algorithm for multi-robot task allocation of the present invention;

[0081] Figure 2 It is the schematic diagram of the representation of the solution of the two-stage evolutionary algorithm for multi-robot task allocation of the present invention;

[0082] Figure 3 It is the description of the decoding process of the two-stage evolutionary algorithm for multi-robot task allocation of the present invention;

[0083] Figure 4 It is the schematic diagram of generating offspring using the PMX method used in the present invention;

[0084] Figure 5 It is the schematic diagram of generating offspring using the 2-opt method used in the present invention;

[0085] Figure 6 It is the schematic diagram of the crowding distance of the solution of density estimation based on displacement used in the present invention. Detailed Embodiments

[0086] The following is a further explanatory description of the present invention in combination with specific embodiments, such as Figure 1As shown in the figure, this embodiment is a multi-robot task allocation method for nuclear accident rescue with two-stage evolution. Its purpose is to allocate task goals for the entire robot cluster given scenario information such as robots, task points, and entrances and exits, as well as conditions of robot types, their quantities, and task types and their quantities. It mainly includes the following steps:

[0087] S1. Initialize multi-robot and task information: Set the number of robots, the number of tasks, and the number of radiation sources. Set information such as the quantities of different types of robots and tasks and their execution efficiencies, and obtain the initial positions of robots and task points, as well as the positions of terrain entrances and exits.

[0088] By initializing multi-robot and task information, a detailed task allocation model can be constructed. The model in this embodiment considers various influencing factors, including:

[0089] Heterogeneity of robots and tasks: Different robots have different functions and performance parameters, and have different efficiencies when performing different tasks, and are suitable for performing different types of tasks; the complexity and urgency of tasks are also different, requiring robots to have different capabilities.

[0090] Multi-objective optimization: Task allocation not only needs to minimize the total task execution time, but also needs to consider multiple objectives such as radiation accumulation and robot waiting time.

[0091] Soft constraint conditions: There are some soft constraint conditions during task execution, such as priorities, time windows, etc., which need to be considered during the optimization process.

[0092] Entrance and exit information: Robots need to enter the task area through specific entrances and leave through specific exits after completing the tasks. The choice of entrance and exit is arbitrary; the goal of multi-robot task allocation is to determine the task execution order of each robot and the corresponding entrance and exit to minimize multiple objective values.

[0093] Then, according to the task allocation scenario at the nuclear accident rescue site, set the scenario instance information and specify the problem values, including the initial state parameters of the robots and the task information in the environment. The scenario instance information includes:

[0094] (1) RN: The number of robots;

[0095] (2) TN: The number and positions of tasks in the area;

[0096] (3) RIN: The number and positions of radiation sources;

[0097] (4) KR: The robot type vector, where each element represents the number of robots of the corresponding type;

[0098] (5) KT: Task type vector, where each element represents the number of tasks of the corresponding type;

[0099] (6) E: Efficiency matrix for different robots to perform different tasks.

[0100] S2. Establishment of optimization problem model: Considering the various cost requirements in the nuclear accident rescue scenario, define the optimization objective, design the encoding of the solution, and represent the solution of the assignment of each task to a robot in an encoded form.

[0101] Use a multi-objective optimization goal, that is, the task assignment not only needs to minimize the total task execution time, but also needs to consider multiple goals such as radiation accumulation and robot waiting time, and assign tasks to each robot according to the criterion of the minimum comprehensive cost.

[0102] In the present invention, consider a multi-robot system I = {1, 2,..., N r}, which contains N r robots; and a task set J = {1, 2,..., N t}, which contains N t tasks, and a set of entrances and exits K = {1, 2,..., N d}.

[0103] In the nuclear rescue scenario considered in the present invention, each robot can only perform one task at a time, and each task can only be performed by one robot at a time.

[0104] In this embodiment, the multi-robot task assignment problem in the nuclear accident rescue scenario can be expressed as the following multi-objective optimization problem:

[0105] (1) Execution time: The time required for a robot to complete a task, affected by the efficiency of the robot; the execution time considers two aspects at the same time: minimizing the total execution time and ensuring a balanced distribution of the execution time among each robot.

[0106] (2) Radiation accumulation: The total radiation exposure endured by a robot during the execution of a task; the goal of radiation accumulation is to minimize the maximum radiation accumulation among all robots.

[0107] (3) Waiting cost: The idle time of a robot when waiting to execute a task or enter a door; the goal of the waiting cost is to minimize the overall task waiting cost.

[0108] The purpose of the optimization objective function is to take into account that some tasks in nuclear rescue have high priorities, and ignoring or delaying the execution of these tasks is likely to lead to the occurrence of secondary accidents.

[0109] Design the optimization objectives to include the robot task execution time, the accumulated radiation received, and the waiting cost, and calculate the optimization objectives; the objective function of the problem is expressed as:

[0110]

[0111]

[0112] Among them, f 1 The goal is to minimize the longest execution time of all robots, f 2 The goal is to minimize the maximum radiation accumulation among all robots, f 3 The goal is to minimize the overall task waiting cost; L k represents the task execution duration of the robot, R k represents the accumulated radiation received by the robot, S k represents the waiting cost of the robot, and I represents the set of all robots in the system.

[0113] The specific process of calculating the optimization goal includes:

[0114] Calculation of the task execution duration of the robot:

[0115] Given the task execution order P of robot k k , the execution time L of the robot k is expressed as:

[0116]

[0117] Among them, represents the task set, is a decision variable determined by the task execution order P k ; when robot k needs to consecutively execute task i and task j, is 1, otherwise it is 0; d ij represents the distance between task points, and represent the distances of the task points from the entrance and the exit, represents the efficiency of robot k in executing task i;

[0118] This formula represents the commuting time spent by the robot entering the scene from the entrance, moving according to the task sequence, and exiting the scene from the exit after the task is completed, plus the total time spent by the robot in executing the tasks;

[0119] Calculation of the accumulated radiation received by the robot:

[0120] Given the average radiation intensity r on each path ij , calculate the radiation accumulated by the robot after passing through this path:

[0121]

[0122] This model describes the cumulative amount of radiation that a robot will receive when passing through a given path with the known positions of radiation sources; there are N s radioactive sources in a given scenario, and their positions are known. Calculate the average radiation intensity r ij :

[0123]

[0124] where and represent the radiation intensities formed by radiation source k at the start and end points of the path respectively; the formula for the variation of radiation from a radiation source with distance usually follows the law of radiation attenuation, and the variation of the radiation intensity of a ray with distance conforms to an exponential attenuation relationship, expressed as:

[0125] I(d) = I 0 ·exp(-μ·d)

[0126] where I(d) represents the radiation intensity at a distance of d, I 0 is the radiation intensity at zero distance (i.e., at the radiation source), and μ is the attenuation coefficient of gamma rays;

[0127] Calculation of the robot's waiting cost:

[0128] Given the urgency coefficient of each task executed by robot k then the waiting cost of robot k can be expressed as:

[0129]

[0130] where ts i is the start execution time of task i, obtained by calculating the total time of the tasks executed before task i; this total time includes the sum of the task execution time and the transmission time required to reach the end point of the task:

[0131]

[0132] where, P k is the task execution sequence of robot k, the efficiency of robot k in executing task , represents the distance between task and task .

[0133] Soft constraint conditions: There are some soft constraint conditions during the task execution process, such as priorities, time windows, etc., which need to be considered during the optimization process.

[0134] The following formula represents the soft constraints of the problem:

[0135]

[0136] This constraint ensures that all tasks will be executed and there are no unassigned tasks.

[0137]

[0138] This constraint guarantees that each task can only be executed by a single robot once, avoiding any task from being executed repeatedly.

[0139]

[0140] This constraint enforces that each task can only be assigned to one robot, which is consistent with the assumptions of single-robot tasks and single-task robots.

[0141]

[0142] This constraint defines the valid range of variables, sets bounds on their values to ensure they are within an acceptable feasible range.

[0143] As Figure 2 shown, in this embodiment, a coding method based on a task assignment matrix is adopted to accurately represent the assignment relationship between robots and tasks:

[0144] The solution is represented by two components, including a task sequence vector and an entry and exit matrix;

[0145] The task sequence vector includes two basic pieces of information: which tasks each robot executes and the execution order of these tasks; in this embodiment, a single-chromosome technique is adopted to represent the task sequence vector, and the length of the vector is N t +N r -1, where N t is the number of tasks and N r is the number of robots.

[0146] Each element in the vector is a unique integer, ranging from [1, N t +N r -1], and there are no duplicate elements; elements with values greater than N r represent separators in the sequence, effectively dividing it into binary subsequences; each subsequence corresponds to the execution sequence of a specific robot; in this embodiment, Figure 2 the task sequence vector in

[0147] represents that there are 3 robots and tasks, the vector length is 5, and elements with values exceeding 3 are designated as separators. t In this embodiment, robot 1 executes task 3, robot 2 executes task 2, and robot 3 executes task 1; the size of the entry and exit matrix is 2*N t, where the vertical data represents the entrance and exit information, and the horizontal data represents the robot ID; the elements in the matrix are integers representing the ID of the corresponding entrance.

[0148] Figure 3 It represents the solution decoding process when N = 7 and M = 3, and this process includes three steps:

[0149] (1) Identification of split elements: First, identify the split elements in the task sequence vector, that is, the elements whose values are greater than 7 (N t ). These elements act as delimiters, dividing the task sequence vector into several subsequences.

[0150] (2) Division of the task sequence vector: Then divide the task sequence vector into N r subsequences according to the split elements, and each subsequence corresponds to the execution sequence of a specific robot.

[0151] (3) Formation of the execution sequence: Finally, combine each subsequence with the corresponding information in the entrance and exit matrix to form the execution sequence of each robot. The entrance and exit matrix provides the detailed information required for the entrance and exit points of the robot during the execution process.

[0152] In this embodiment, robot 1 enters the working area from port 4, sequentially executes tasks numbered 5 and 7, and finally exits from port 1.

[0153] S3. Multi-robot task allocation: Use a two-stage evolutionary algorithm to optimize the task allocation scheme around the optimization objective in step S2 to obtain a set of solutions for the optimal task allocation scheme: As Figure 1 shown, the specific process of the two-stage evolutionary algorithm in this embodiment includes:

[0154] S31. Initialize the population: Initialize the population P, set the population size to N, and initialize the function evaluation number FE of the algorithm to N; ToSEA uses the number of function evaluations as its termination condition, and the maximum number of function evaluations FE max can be set according to specific situations.

[0155] S32. Generate offspring solutions: Calculate the stage parameters; for the parental population in step S31, obtain the offspring solutions Q according to different offspring generation methods in different stages.

[0156] The specific generation process of the offspring solutions:

[0157] Calculate the stage parameter I representing the current loop stage stage :

[0158]

[0159] where, r cis a parameter for controlling the two - level ratio; FE represents the number of function evaluations performed by the current algorithm; FE max represents the maximum allowed number of function evaluations.

[0160] According to different stage parameters, different offspring generation methods are used in different stages; in the initial stage of the search process, the solutions of the algorithm have not converged to the Pareto front of the problem, and the focus of the algorithm at this stage is to enhance the convergence of the solutions.

[0161] As the algorithm enters the second stage, the solutions tend to converge or approach the Pareto front of the problem. Therefore, local search is performed on the obtained solutions to improve the diversity of the solution set.

[0162] When I stage = 1:

[0163] This stage is in the initial stage of the search. The solutions of the algorithm have not converged to the Pareto front of the problem. The focus of the algorithm in this stage is to enhance the convergence of the solutions; the generation of offspring follows the principles of a typical evolutionary algorithm. First, the binary tournament selection method is used to obtain the hybridization pool, and then offspring are generated based on genetic operators; in the binary tournament selection, the fitness of an individual is determined by its rank in the non - dominated sorting and crowding distance;

[0164] As Figure 4 shown, since the solutions are encoded in sequence form, the partially - matched crossover (PMX) method is used to generate offspring, where the entry and exit matrices of the two child nodes are inherited from the parent nodes.

[0165] The mutation operation is set to swap two different elements selected from the task sequence vector, and the probability of randomly regenerating the entry and exit matrices is 0.2; the focus of this stage is to enhance the convergence of the solutions, so that the solutions of the algorithm quickly converge to the Pareto front of the problem.

[0166] When I stage = 2:

[0167] Local search is performed on the solution with the largest crowding distance in the population, and duplicate solutions are excluded. Then, the neighbors obtained from this search are used as newly generated offspring, and the solutions are added to the exploration set after the local search is completed; in this stage, the search set E is used to track the solutions that have undergone local search;

[0168] As Figure 5 shown, the 2 - opt method is used for local search to quickly obtain improved solutions; only the task sequence vector is operated in this stage, while keeping the entry and exit matrices unchanged; in the second stage, the solutions tend to converge or approach the Pareto front of the problem, and at this time, the local search performed on the population can improve the diversity of the solution set.

[0169] S33. Environmental Selection: Combine the parental population and the offspring, calculate the individual fitness according to the stage parameters obtained in step 32, and screen the individuals using different screening methods based on the individual fitness.

[0170] According to the different stage parameters, the environmental selection process is divided into two parts:

[0171] When I stage = 1:

[0172] First, normalize the objective values of the solution set to reduce the impact of different objective scales on the search process; then perform non-dominated sorting on the population, and incorporate each rank into the next-generation population in ascending order, starting from the lowest rank until adding the last rank exceeds the population size constraint N;

[0173] Then, according to the fitness of each solution in the last rank, remove the solution with the maximum fitness one by one until the population size is reduced to N; during this process, the fitness needs to be recalculated each time a solution is excluded;

[0174] Calculate the fitness of the solution according to its non-dominated rank and crowding distance:

[0175]

[0176] Among them, R represents the non-dominated sorting rank, and C represents the crowding distance; the lower the fitness value, the higher the quality of the solution.

[0177] Preferentially select the solution with the lowest non-dominated rank. When all solutions are in the same rank, the crowding distance becomes the decisive factor determining its fitness; the crowding distance of the solution is calculated using the displacement-based density estimation (SDE) method, as Figure 6 shown. When calculating the crowding distance of the current solution using the SDE method, transfer other solutions to the boundary of the dominant region of the current solution; the calculation of the transfer point is shown in the following formula,

[0178]

[0179] where q′ is the target value point corresponding to the transfer point, and p is the target point of the solution for which the crowding distance is calculated;

[0180] When I stage = 2:

[0181] The algorithm cancels the population size limit and deletes the dominant and duplicate solutions in the population.

[0182] S34. Loop Output: Compare whether the function evaluation times FE reach the maximum value FE max , if it reaches the maximum value, output the final population solution set, otherwise return to step S32 to continue the next loop.

[0183] As described above, it is only a further explanatory illustration of the present invention in combination with specific embodiments. All the descriptions made do not represent a limitation on the protection scope of the present invention. Any changes or alternative solutions that can be easily conceived by any person skilled in the art within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the said claims.

Claims

1. A two-stage evolutionary method for allocating tasks of multiple robots in nuclear accident rescue, comprising the following steps: S1. Initialize multi-robot and task information: set the number of robots, tasks and radiation sources, set the number of different types of robots and tasks and execution efficiency, obtain the initial position of the robot and task point, and the location of the terrain entrance and exit, etc.; set the scenario instance information according to the task allocation scenario of the nuclear accident rescue site, and specify the problem value; S2. Establishment of optimization problem model: Comprehensively consider the various cost requirements in the nuclear accident rescue scenario, define the optimization goal, design the solution encoding, and represent the solution of each task and robot allocation in a coding form; S3. Multi-robot task allocation: Use a two-stage evolutionary algorithm to optimize the task allocation plan around the optimization goal in step S2 to obtain a set of optimal task allocation solution solutions.

2. The two-stage evolutionary algorithm for multi-robot task allocation in nuclear accident rescue according to claim 1 is characterized by: The scene instance information described in step S1 initializing the multi-robot and task information includes: (1) RN: number of robots; (2) TN: number and location of tasks in the region; (3) RIN: number and location of radiation sources; (4) KR: robot type vector, where each element represents the number of robots of the corresponding type; (5) KT: task type vector, where each element represents the number of tasks of the corresponding type; (6) E: Efficiency matrix of different robots performing different tasks.

3. The two-stage evolutionary algorithm for multi-robot task allocation in nuclear accident rescue according to claim 2 is characterized by: The specific process of defining the optimization target in the optimization problem model establishment in step S2 includes: Designing optimization objectives including robot task execution time, accumulated radiation exposure and waiting cost, and calculating the optimization objectives; The objective function of the problem is expressed as: minf1=max k∈I L k minf2=max k∈I R k Among them, the f1 goal is to minimize the longest execution time of all robots, the f2 goal is to minimize the maximum radiation accumulation between all robots, and the F3 goal is to minimize the overall task waiting cost; L k Indicates the robot task execution time, R k Indicates the accumulated radiation received by the robot, S k represents the waiting cost of the robot, and I represents the set of all robots in the system.

4. The two-stage evolutionary algorithm for multi-robot task allocation in nuclear accident rescue according to claim 3 is characterized by: The specific process of calculating the optimization target includes: Calculation of robot task execution time: Given the task execution order P of robot k k , the robot's execution time L k It is expressed as: in, Indicates a collection of tasks. is the decision variable, which is determined by the task execution order P k When robot k needs to perform tasks i and j in succession, is 1, otherwise it is 0; d ij represents the distance between task points, and Indicates the distance between the task point and the entrance and exit. represents the efficiency of robot k in performing task i; This formula represents the total time the robot spends entering the scene from the entrance, moving according to the task sequence, and exiting the scene from the exit after the task is completed, plus the total time the robot spends performing the task; Calculation of the accumulated radiation to which the robot is exposed: Given the average radiation intensity r on each path ij , calculate the accumulated radiation after the robot passes through the path: The model describes the cumulative amount of radiation that the robot will receive when passing through a given path when the location of the radiation source is known. s The radiation source, and the position is known, calculate the average radiation intensity r ij : in, and They represent the radiation intensity formed by the radiation source k at the starting point and the end point of the path respectively; the formula for the change of radiation source radiation with distance usually follows the law of radiation attenuation, and the change of ray radiation intensity with distance conforms to the exponential attenuation relationship, which can be expressed as: I(d)=I0·exp(-μ·d) Where I(d) represents the radiation intensity at a distance of d, I0 is the radiation intensity at zero distance (i.e., at the radiation source), and μ is the attenuation coefficient of gamma rays; Calculation of robot waiting cost: Given the urgency coefficient of each task performed by robot k Then the waiting cost of robot k can be expressed as: where ts i is the start execution time of task i, which is obtained by calculating the total time of the tasks executed before task i; this total time includes the sum of the task execution time and the transmission time required to reach the task end point: Among them, P k is the task execution sequence of robot k, Robot K performs tasks efficiency, Indicates the task and tasks The distance between.

5. The two-stage evolutionary algorithm for multi-robot task allocation in nuclear accident rescue according to claim 4 is characterized by: The encoding of the solution in the optimization problem model establishment in step S2 adopts an encoding method based on the task allocation matrix to accurately represent the allocation relationship between the robot and the task: The solution is represented by two components, including a task order vector and an entry and exit matrix; The task order vector includes two basic information: which tasks each robot performs and the order in which these tasks are performed; The single chromosome technology is used to represent the task sequence vector, the length of the vector is N t +N r -1, where N t is the number of tasks, N r is the number of robots.

6. The two-stage evolutionary algorithm for multi-robot task allocation in nuclear accident rescue according to claim 1 is characterized by: The specific process of step S3 multi-robot task allocation includes: S31. Initialize the population: Generate an initial population so that each individual in the population represents a possible task allocation scheme, and set the population size and loop termination conditions; S32, generating offspring solutions: calculating stage parameters; for the initial population in step S31, selecting different offspring generation methods according to different stages to obtain offspring solutions; S33, environment selection: merge the parents and offspring, calculate the individual fitness according to the stage parameters obtained in step S32, and screen the individuals using different screening methods according to the individual fitness; S34, loop output: check whether the loop termination condition is met, if so, output the final population cluster solution, otherwise return to step S32.

7. The two-stage evolutionary algorithm for multi-robot task allocation in nuclear accident rescue according to claim 6 is characterized by: The loop termination condition in the initialization population in step S31 is: Calculate the number of function evaluations FE = N in the loop, where N is the population size; when the number of function evaluations reaches the maximum FE max The loop terminates.

8. The two-stage evolutionary algorithm for multi-robot task allocation in nuclear accident rescue according to claim 7 is characterized by: The specific process of obtaining the child solution in step S32 includes: Calculate the phase parameter I representing the current cycle phase stage : Among them, r c It is a parameter that controls the ratio of the two levels; FE represents the number of function evaluations executed by the current algorithm; FE max Indicates the maximum allowed number of function evaluations; Depending on the stage parameters, different child generation methods are used in different stages: When I stage =1: The generation of offspring follows the principles of typical evolutionary algorithms. First, a hybrid pool is obtained using a binary competition selection method, and then offspring are generated based on genetic operators. In the binary competition selection, the fitness of an individual is determined by its rank and crowding degree in the non-dominated sorting. The offspring are generated using the partial matching crossover PMX method, where the entry and exit matrices of the two child nodes are inherited from the parent node; The mutation operation is set to select two different elements from the task sequence vector for exchange, and the probability of randomly regenerating the entry and exit matrices is 0.2; When I stage =2: Perform a local search for the solution with the largest crowding distance within the population and exclude duplicate solutions. Then use the neighbors obtained from the search as the newly generated offspring. After completing the local search, add the solution to the exploration set. In this stage, use the search set E to track the solutions that have been locally searched. The local search uses the 2-opt method to quickly obtain an improved solution; in this stage, only the task order vector is operated, while the entry and exit matrices are kept unchanged.

9. The two-stage evolutionary algorithm for multi-robot task allocation in nuclear accident rescue according to claim 8, characterized in that: The specific process of the environment selection in step S33 includes: Depending on the stage parameters, the environment selection process is divided into two parts: When I stage =1: First, the target value of the solution set is normalized, and then the population is non-dominated sorted, and each order is included in the next generation population in ascending order, starting from the lowest order, until the last order is added and exceeds the population size constraint N; Then, according to the fitness of each solution in the last stage, the solutions with the largest fitness are removed one by one until the population size is reduced to N; In this process, the fitness needs to be recalculated every time a solution is eliminated; Calculate the fitness of a solution based on its nondominated rank and crowding distance: Among them, R represents the non-dominated sorting level, and C represents the crowding distance. The lower the fitness value, the higher the quality of the solution. The crowding distance of the solution is calculated using the displacement-based density estimation SDE method. When calculating the crowding distance of the current solution, other solutions are transferred to the boundary of the dominant area of ​​the current solution. The calculation of the transfer point is shown in the following formula: Where q′ is the target value point corresponding to the transfer point, and p is the target point of the solution for calculating the crowding distance; When I stage =2: The algorithm removes the population size restriction and deletes the dominant and duplicate solutions in the population.

10. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, any step of the two-stage evolutionary algorithm for multi-robot task allocation in nuclear accident rescue as described in any one of claims 1-9 is implemented.

Images