A multi-robot control valve assembly task allocation optimization method based on a hybrid GA algorithm

Abstract

The application discloses a kind of based on the optimization method of mixed GA algorithm's multi-robot control valve assembly task allocation, applied to multi-robot control valve assembly system, comprising: initialization control valve assembly task data: determine the type of control valve to be assembled, the assembly quantity of each type of control valve to be assembled, and the assembly step of each type of control valve to be assembled;According to the control valve assembly task data after initialization, obtain the total assembly step of all types of control valve to be assembled as total task number;According to preset parameter, based on total task number, using mixed GA algorithm to iteratively optimize the task allocation scheme of control valve assembly until reaching preset maximum iteration number, select the individual with the highest fitness in record as the best allocation scheme.The method is based on mixed GA algorithm, in complex and changeable production environment, realize the efficient, reasonable allocation of various control valve assembly tasks in multi-robot system.

Description

Technical Field

[0001] This invention relates to the fields of intelligent equipment technology, multi-robot collaboration technology, and autonomous assembly task allocation technology, and more specifically to a multi-robot control valve assembly task allocation optimization method based on a hybrid GA algorithm. Background Technology

[0002] In modern industrial production, especially in control valve assembly, multi-robot collaboration to complete complex tasks has become a key trend for improving production efficiency and flexibility. However, how to achieve efficient and dynamic task allocation, enabling robots to work collaboratively while ensuring load balancing and coping with various uncertainties in the industrial environment, remains a core research challenge.

[0003] Existing task allocation algorithms mainly include heuristic rule-based methods and simple genetic algorithms. These methods often struggle to achieve ideal dynamic scheduling and load balancing when handling complex tasks, especially in industrial environments where external factors such as hardware failures, network latency, and unstable communication significantly increase the complexity of task allocation. These uncertainties not only affect the real-time performance and accuracy of task allocation but also limit the performance of existing algorithms in practical applications.

[0004] The Hybrid Enhanced Genetic Algorithm (GA) is an optimization algorithm based on natural selection and genetic mechanisms, widely used for optimizing task allocation in complex tasks. This algorithm significantly improves the efficiency and robustness of task allocation through the integration of several key technologies. Specifically, Hybrid GA introduces a heuristic initialization strategy to accelerate the generation of initial solutions; employs a vectorized computation method to improve computational efficiency; combines an adaptive non-uniform mutation mechanism to enhance the algorithm's exploration capability in the search space; and incorporates local search techniques to further optimize the accuracy of solutions. Furthermore, Hybrid GA fully considers external factors in industrial environments, such as hardware failure rates, network latency, and communication stability, enabling the task allocation scheme to dynamically adapt to environmental changes and ensuring the system maintains high efficiency and stability even in the face of uncertainty.

[0005] Therefore, how to efficiently and rationally allocate the assembly tasks of various control valves in a multi-robot system based on the hybrid GA algorithm in a complex and ever-changing production environment is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of the above problems, the present invention provides a multi-robot control valve assembly task allocation optimization method based on a hybrid GA algorithm, so as to at least solve some of the technical problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] This invention provides a multi-robot control valve assembly task allocation optimization method based on a hybrid GA algorithm, applicable to a multi-robot control valve assembly system, comprising the following steps:

[0009] Initialize control valve assembly task data: determine the type of control valve to be assembled, the assembly quantity of each type of control valve, and the assembly steps for each type of control valve.

[0010] Based on the initialized control valve assembly task data, obtain the total number of assembly steps for all types of control valves to be assembled, which is then used as the total number of tasks.

[0011] Based on preset parameters and the total number of tasks, a hybrid GA algorithm is used to iteratively optimize the task allocation scheme for control valve assembly, specifically including:

[0012] (1) Initialize the population based on the total number of tasks, where each individual in the population represents a task allocation scheme;

[0013] (2) Obtain the fitness and load balance of each individual in the current iteration, and record the individual with the best fitness and its corresponding load balance.

[0014] (3) Determine if each task has timed out and reallocate the timed-out tasks;

[0015] (4) Based on the preset number of elites to be retained, an elite retention strategy is adopted, and individuals are retained in descending order of fitness, so as to retain multiple elite individuals with the best fitness.

[0016] (5) Select, crossover, and mutate non-elite individuals to generate a new population;

[0017] (6) Perform local search optimization on the new population;

[0018] (7) Merge the elite individuals and the optimized new population to generate a new generation of population;

[0019] (8) Based on the new generation population, repeat the above steps (2)-(7) until the preset maximum number of iterations is reached, and select the individual with the highest fitness in the record as the best allocation scheme.

[0020] Furthermore, it also includes:

[0021] Based on the optimal allocation scheme and relevant statistical data, draw a fitness history chart, a load balancing history chart, and a task allocation chart.

[0022] Furthermore, the total number of tasks is expressed as:

[0023]

[0024] Where T represents the total number of assembly steps for all types of control valves to be assembled, i.e., the total number of tasks; V represents the total number of V types of control valves to be assembled; N v S represents the assembly quantity of the control valves of type v to be assembled; v This indicates the assembly steps for the control valve of type v to be assembled.

[0025] Furthermore, the initialization of the population based on the total number of tasks specifically includes:

[0026] Create a zero matrix of size equal to the population size multiplied by the total number of tasks; where the population size is the number of drones.

[0027] Each individual is initialized using a parallel loop;

[0028] Randomly arrange the task order and assign the tasks sequentially to the robot with the lowest current load to obtain the initial population matrix;

[0029] By introducing 10% random mutation, the initial population matrix is ​​optimized to obtain the final initialized population.

[0030] Furthermore, the fitness, which is the maximum total completion time of all robot tasks, is expressed as:

[0031]

[0032] Where H represents the maximum total completion time of all robot tasks, i.e., fitness; R represents the total number of robots; T represents the total number of tasks; t i Indicates the execution time of task i; δ i,r Let δ represent the task assignment indicator function. If task i is assigned to robot r, then δ i,r It is 1 if it is true, otherwise it is 0.

[0033] Furthermore, the load balancing degree is expressed as:

[0034]

[0035]

[0036] Where B represents load balancing degree; T r σ(T) represents the total task time of robot r; r ) represents T rStandard deviation; μ(T) r ) represents T r The mean of the ; Q represents a constant to prevent division by zero; R represents the total number of robots.

[0037] Furthermore, step (3) specifically includes:

[0038] Obtain the base assembly time for each task;

[0039] Based on the base assembly time for each task, obtain the task assembly time after considering the hardware failure rate; expressed as:

[0040] t' base =t base ×(1+0.5×F)

[0041] F~Bernoulli(P f )

[0042] Among them, t' base This represents the task assembly time after considering hardware failure rate; t base Represents the base assembly time of the task; F represents the hardware fault indication variable; P f This represents the hardware failure rate; Bernoulli represents the Bernoulli distribution.

[0043] Based on the task assembly time taking into account hardware failure rate, and combined with network latency and communication overhead, the actual execution time of each task is obtained; expressed as:

[0044] t = max(([t' base ×(1+0.5×F)])+D+C,0.1)×(1+ε)

[0045] Where t represents the actual execution time of a single task; D represents network latency; C represents communication overhead; ε represents the fluctuation factor, and ε~η(0, 0.02) 2 ), indicating that the fluctuation factor ε follows a normal distribution with a mean of 0 and a standard deviation of 0.02; max(·) ensures that the task time is not less than 0.1 seconds;

[0046] in, This indicates that the network latency D follows a mean of μ. d The standard deviation is σ d The normal distribution; This represents a normal distribution; and D = max(D, 0);

[0047] The communication overhead C is represented as:

[0048]

[0049] If communication is unstable, add a fixed 5 seconds; otherwise, do not add any.

[0050] Determine whether the actual execution time of each task exceeds the preset time. If there are timed-out tasks, trigger the retry mechanism to reassign the timed-out tasks to other robots.

[0051] Furthermore, step (5) specifically includes:

[0052] The tournament selection method is used to select non-elite individuals;

[0053] A partial matching crossover method is used to perform crossover operations on selected non-elite individuals to generate offspring;

[0054] A new population is generated by adjusting the genes that produce offspring through adaptive non-uniform mutation.

[0055] Furthermore, step (6) specifically includes:

[0056] Based on the total task time of each robot, identify the robot with the heaviest workload;

[0057] Determine if the robot with the heaviest load is overloaded.

[0058] If there are overloaded tasks, the task with the longest execution time that the robot is responsible for will be selected and the task will be reassigned to the robot with the lightest load.

[0059] The simulated annealing algorithm is used to optimize the new task allocation and obtain an optimized new population.

[0060] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a multi-robot control valve assembly task allocation optimization method based on a hybrid GA algorithm, which has the following beneficial effects:

[0061] This invention achieves highly efficient optimization of task allocation in multi-robot systems by enhancing a hybrid genetic algorithm and comprehensively considering various external factors in the industrial environment. Specific advantages include:

[0062] (1) Comprehensive consideration of multiple factors: External factors such as hardware failure, network latency and communication stability are introduced during the task time generation process to make the optimization results closer to the actual industrial environment.

[0063] (2) Enhancement and hybridization of optimization algorithms: Combining genetic algorithms with hybrid local search (greedy algorithm and simulated annealing) improves the search ability and optimization effect of the algorithm and avoids getting trapped in local optima.

[0064] (3) Dynamic adaptation mechanism: By dynamically adjusting the mutation rate and the number of elites retained, the diversity of the population is maintained, and the robustness and adaptability of the algorithm are improved.

[0065] (4) Parallel computing acceleration: By utilizing parallel computing technology, the execution efficiency of the algorithm is improved, and the needs of large-scale task allocation are met.

[0066] (5) Comprehensive results presentation: Through various visualization methods, it provides intuitive optimization process and results analysis to assist decision-makers in making effective evaluations and adjustments.

[0067] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0068] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0069] Figure 1 This is a schematic diagram of the multi-robot control valve assembly task allocation optimization method based on the hybrid GA algorithm provided in an embodiment of the present invention.

[0070] Figure 2 This is a schematic diagram of the assembly task allocation optimization process provided in an embodiment of the present invention.

[0071] Figure 3 This is a schematic diagram illustrating the optimal allocation scheme provided in an embodiment of the present invention.

[0072] Figure 4 This is a schematic diagram of the visual content provided for an embodiment of the present invention.

[0073] Figure 5 This is a schematic diagram of task allocation provided in an embodiment of the present invention. Detailed Implementation

[0074] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0075] This invention discloses an optimization method for multi-robot control valve assembly task allocation based on a hybrid GA algorithm. See [link to relevant documentation]. Figure 1 As shown, it includes the following steps:

[0076] S1. Initialize control valve assembly task data: Determine the type of control valve to be assembled, the assembly quantity of each type of control valve, and the assembly steps for each type of control valve.

[0077] S2. Based on the initialized control valve assembly task data, obtain the total assembly steps for all types of control valves to be assembled, and use this as the total number of tasks.

[0078] S3. Based on the total number of tasks and according to the preset parameters, the task allocation scheme for the control valve assembly is optimized using a hybrid GA algorithm to obtain the best allocation scheme.

[0079] S4. Based on the optimal allocation scheme and relevant statistical data, draw the fitness history chart, load balancing history chart, and task allocation chart.

[0080] In step S1 above, the control valve assembly task data is initialized: the type of control valve to be assembled, the assembly quantity of each type of control valve, and the assembly steps of each type of control valve are determined.

[0081] For the assembly steps of the control valve to be assembled, a complete assembly is divided into several smaller parts according to the interference situation. Each part is called a sub-assembly of the assembly, and each sub-assembly corresponds to an assembly step, that is, an assembly task.

[0082] In step S2 above, based on the initialized control valve assembly task data, the total number of assembly steps for all types of control valves to be assembled is obtained, which is taken as the total number of tasks; this total number of tasks is expressed as:

[0083]

[0084] Where T represents the total number of assembly steps for all types of control valves to be assembled, i.e., the total number of tasks; V represents the total number of V types of control valves to be assembled; N v S represents the assembly quantity of the control valves of type v to be assembled; v This indicates the assembly steps for the control valve of type v to be assembled.

[0085] In step S3 above, based on preset parameters and the total number of tasks, a hybrid GA algorithm is used to optimize the task allocation scheme for control valve assembly, obtaining the optimal allocation scheme; see [link to relevant documentation]. Figure 2 As shown, it specifically includes:

[0086] (1) Initialize the population based on the total number of tasks, where each individual in the population represents a task allocation scheme;

[0087] Population initialization in this step has the following effects: ① Ensuring population diversity: Randomly generating initial individuals covers the search space as much as possible, preventing the algorithm from converging to a local optimum prematurely. ② Providing a foundation for evolution: Proper initialization ensures that subsequent crossover, mutation, and other operations can play an effective role. ③ Task load balancing: In task allocation problems, the initialization process should distribute tasks to the robot as evenly as possible, providing a good starting point for subsequent optimization.

[0088] The specific implementation process includes:

[0089] ① Create a zero matrix of size equal to the population size multiplied by the total number of tasks; where the population size is the number of drones; assuming there are 5 robots (i.e., the population size is 5) and 10 different assembly tasks need to be completed (i.e., the total number of tasks is 10); then a 5x10 zero matrix will be created.

[0090] ② Initialize each individual using a parallel loop; assign tasks to each individual;

[0091] For example, for the 5x10 matrix above, 5 individuals need to be initialized. Each individual is a task allocation scheme of length 10, indicating which robot will perform each of the 10 tasks. For example, a possible initialization scheme can be represented as:

[0092] Individual 1: [Robot 1, Robot 2, Robot 3, Robot 4, Robot 5, Robot 1, Robot 2, Robot 3, Robot 4, Robot 5];

[0093] Individual 2: [Robot 2, Robot 3, Robot 4, Robot 5, Robot 1, Robot 2, Robot 3, Robot 4, Robot 5, Robot 1];

[0094] Individual 3: [Robot 3, Robot 4, Robot 5, Robot 1, Robot 2, Robot 3, Robot 4, Robot 5, Robot 1, Robot 2];

[0095] Individual 4: [Robot 4, Robot 5, Robot 1, Robot 2, Robot 3, Robot 4, Robot 5, Robot 1, Robot 2, Robot 3];

[0096] Individual 5: [Robot 5, Robot 1, Robot 2, Robot 3, Robot 4, Robot 5, Robot 1, Robot 2, Robot 3, Robot 4];

[0097] ③ Randomly arrange the task order and assign the tasks sequentially to the robot with the lowest current load to obtain the initial population matrix; specifically:

[0098] To ensure the diversity of the initial population, the tasks are first randomly sorted. Assume the randomly sorted task list is: [Task 7, Task 2, Task 9, Task 1, Task 6, Task 8, Task 3, Task 5, Task 4, Task 10]. Then, the tasks are assigned sequentially to the robot with the lowest current workload according to the order in the list. The task assignment rule is as follows:

[0099]

[0100] Where r represents the selected robot number, i.e., the r-th robot; R represents the total number of robots; L o This represents the load of robot o; the argmin operator is used to find L. o Minimize the value of o.

[0101] For each task to be assigned, the algorithm iterates through all robots and selects the robot r with the lowest current load to execute the task. This ensures load balancing in task allocation and prevents any robot from becoming a bottleneck due to excessive tasks.

[0102] Finally, update the current load of each robot, as shown below:

[0103] L r ←L r +t k

[0104] Among them, t k L represents the execution time of task k; r This indicates the current load of the selected robot; once a task is assigned to a robot, the robot's load L... r This will increase t k The load is measured in seconds to reflect the time required for a newly assigned task. (Load is the time remaining for the robot to complete its current task, measured in seconds.) This step ensures that load information is updated in real time, providing accurate load data for subsequent task assignments.

[0105] ④ Introduce 10% random mutation to optimize the initial population matrix and obtain the final initialized population; represented as:

[0106]

[0107] Where pm represents the mutation probability; P k The k-th gene of an individual is represented; Randi represents matrix operations; Randi(1, m) represents generating a random integer between 1 and m, representing a randomly selected robot number.

[0108] For each gene (task assignment) of each individual in the population, mutation is performed with a 10% probability. The mutation operation replaces the current gene value pk with a randomly selected robot number Randi(1,m). This random mutation helps increase the diversity of the population and avoids the genetic algorithm getting trapped in local optima.

[0109] (2) Obtain the fitness and load balance of each individual in the current iteration, and record the individual with the best fitness and its corresponding load balance; where:

[0110] Fitness, which is the maximum total time to complete all robot tasks, is expressed as:

[0111]

[0112] Where H represents the maximum total completion time of all robot tasks, i.e., fitness; R represents the total number of robots; T represents the total number of tasks; t i Indicates the execution time of task i; δ i,r Let δ represent the task assignment indicator function. If task i is assigned to robot r, then δ i,r It is 1 if it is true, otherwise it is 0.

[0113] Load balancing is expressed as:

[0114]

[0115] Where B represents load balancing degree; T r σ(T) represents the total task time of robot r; r ) represents T r Standard deviation; μ(T) r ) represents T r The mean of the load balance is denoted by Q; Q represents a constant to prevent division by zero; R represents the total number of robots. A smaller load balance value indicates a more balanced load.

[0116] (3) Determine if each task has timed out and reassign timed-out tasks; specifically including:

[0117] ① Obtain the base assembly time for each task;

[0118] ② Based on the base assembly time for each task, obtain the task assembly time after considering the hardware failure rate; expressed as:

[0119] t' base =t base ×(1+0.5×F)F~Bernoulli(P f )

[0120] Among them, t' baseThis represents the task assembly time for a single task, taking into account hardware failure rate; t base Represents the base assembly time for a single task; F represents a hardware fault indicator variable, where F = 1 if a fault occurs, otherwise F = 0; P f This represents the hardware failure rate; Bernoulli represents the Bernoulli distribution.

[0121] ③ Based on the task assembly time considering hardware failure rate, and combined with network latency and communication overhead, the actual execution time of each task is obtained; expressed as:

[0122] t = max(([t' base ×(1+0.5×F)])+D+C,0.1)×(1+ε)

[0123] Where t represents the actual execution time of a single task; D represents network latency; C represents communication overhead; ε represents the fluctuation factor, and ε~η(0, 0.02) 2 ), indicating that the fluctuation factor ε follows a normal distribution with a mean of 0 and a standard deviation of 0.02; max(·) ensures that the task time is not less than 0.1 seconds;

[0124] in, This indicates that the network latency D follows a mean of μ. d The standard deviation is σ d The normal distribution; This represents a normal distribution; and D = max(D, 0);

[0125] The communication overhead C is represented as:

[0126]

[0127] If communication is unstable, add a fixed 5 seconds; otherwise, do not add any.

[0128] ④ Determine if the actual execution time of each task exceeds the preset time. If a timeout task exists, trigger the retry mechanism to reassign the timeout task to another robot; this prevents a robot from timed out due to excessive task load. It's important to note that each retry must ensure the task is assigned to a new robot, not the original one. Furthermore, to avoid performance issues caused by infinite retries, the system specifies a maximum number of retries per task. If a task cannot be completed within the maximum number of retries, it may be considered a failure.

[0129] (4) Based on the preset number of elites to be retained, an elite retention strategy is adopted, and individuals are retained in descending order of fitness, so as to retain multiple elite individuals with the best fitness.

[0130] For example, if the preset number of elite individuals to be retained is E, that is, the top E individuals are selected as elite individuals based on their fitness from high to low; if the population diversity is below a threshold, the number of elite individuals is increased to maintain superior genes; this can be expressed as:

[0131] C = min(E + ΔE, P)

[0132]

[0133] Where ΔE represents the additional number of elites adjusted according to diversity; θ represents the diversity threshold; P represents the population size; when the population diversity is below the threshold, the number of elites is increased to maintain superior genes.

[0134] (5) Select, crossover, and mutate non-elite individuals to generate a new population; specifically:

[0135] ① The selection process for non-elite individuals using the tournament selection method is represented as follows:

[0136]

[0137] Here, Select(s) represents the parent individual selected for the s-th individual in the new population; arg min represents the index of the individual that minimizes fitness; and Fitness(j) represents the fitness value of individual j. The lower the fitness, the better the individual.

[0138] ② The partial matching crossover method (PMX) is used to perform crossover operations on selected non-elite individuals to generate offspring; specifically, the legitimacy of the gene sequence is maintained, and legitimate offspring are generated by exchanging some genes from the parents.

[0139] ③ The genes that generate offspring are adjusted through adaptive non-uniform mutation to generate a new population; in this process, the mutation probability decreases with the number of iterations; expressed as:

[0140]

[0141] g∈{1, 2, ..., G}

[0142] Among them, P m represents the initial mutation probability; g represents the current iteration number; G represents the maximum iteration number; when dynamic adjustment is enabled, the mutation probability decays exponentially; otherwise, it decays linearly.

[0143] (6) Perform local search optimization on the new population;

[0144] First, a greedy algorithm is used for local search to quickly improve the quality of the current solution; including:

[0145] ① Identify the robot with the heaviest workload based on the total task time of each robot;

[0146] ② Determine if the robot with the heaviest load has an overloaded task; if so, select the task with the longest execution time currently handled by that robot and reassign it to the robot with the lightest load; specifically:

[0147]

[0148] Among them, t * Indicates the robot with the heaviest load identified; T overload For the set of tasks of an overloaded robot; t i Let be the execution time of task i; load(r) is the current load of robot r.

[0149] Subsequently, simulated annealing is used to escape local optima during local search, exploring a wider solution space by accepting a certain probability of poorer solutions. In this embodiment of the invention, after greedily reallocating tasks, the new task allocation is further optimized to reduce the total completion time and obtain an optimized new population. Specifically:

[0150] The probability formula for simulated annealing accepting a new solution is:

[0151]

[0152] Where ΔW represents the energy difference between the new solution and the current solution (and the fitness difference at point Q); Q is 100 here, based on debugging and optimization.

[0153] In this embodiment of the invention, the greedy algorithm first quickly identifies and reallocates tasks for the overloaded robot, and then further optimizes the task allocation through simulated annealing to ensure the quality and diversity of solutions. The greedy algorithm provides rapid local improvement, while simulated annealing enhances the global search capability; the combination of the two effectively avoids getting trapped in local optima. Combined with techniques such as dynamic mutation rate, diversity maintenance, and parallel computing, the overall algorithm can efficiently optimize robot task allocation and improve production efficiency while considering the complexities of the industrial environment.

[0154] (7) Merge the elite individuals and the optimized new population to generate a new generation of population;

[0155] (8) Based on the new generation population, repeat the above steps (2)-(7) until the preset maximum number of iterations is reached, and select the individual with the highest fitness in the record as the best allocation scheme.

[0156] In step S4 above, based on the optimal allocation scheme and relevant statistical data, a fitness history chart, a load balancing history chart, and a task allocation chart are drawn. Specifically, this includes:

[0157] ① Calculate diversity metrics based on fitness and load balancing history;

[0158] ② Generate diversity history vectors for visualization;

[0159] This can be expressed by the formula:

[0160]

[0161] Where D(t) represents the diversity index; α, β, and γ all represent weight parameters; B represents the load balance degree; and μ(t) represents the historical mean of fitness.

[0162] In summary, this invention provides a multi-robot control valve assembly task allocation optimization method based on a hybrid GA algorithm to solve problems such as low task allocation efficiency, unbalanced load, and slow system response in existing technologies. By combining the global search capability of genetic algorithms with the fine optimization of local search algorithms, this method can achieve efficient and reasonable allocation of various control valve assembly tasks in multi-robot systems under complex and ever-changing production environments, mainly in the following aspects:

[0163] (1) Adjustment of dynamic variation rate

[0164] This invention dynamically adjusts the mutation rate based on population diversity and the current generation number to maintain population diversity and fitness. By monitoring population diversity indicators, when the population is found to be becoming homogeneous, the mutation rate is appropriately increased to introduce new genetic information, thereby preventing the algorithm from getting trapped in local optima.

[0165] (2) Integrated cross-local search strategy

[0166] This invention further optimizes the task allocation scheme by combining a greedy strategy and simulated annealing algorithm for local search after a certain number of generations. The greedy strategy is used to quickly find local optima, while the simulated annealing algorithm enhances the global search capability by accepting suboptimal solutions with a certain probability to escape local optima.

[0167] (3) Task timeout and retry mechanism

[0168] This invention introduces a task timeout detection and retry mechanism to ensure effective task reallocation in the event of a timeout. This mechanism monitors task execution progress and automatically triggers a task reallocation process when a timeout is detected, improving system robustness and reliability and preventing the failure of a single task from impacting overall production efficiency.

[0169] (4) Vectorization of fitness calculation

[0170] This invention utilizes parallel computing technology to vectorize the fitness function, accelerating the fitness calculation process. By parallelizing the fitness evaluation of multiple individuals, the computational efficiency of the algorithm is significantly improved, the overall optimization time is shortened, and the needs of high-complexity task allocation are met.

[0171] (5) Population diversity maintenance

[0172] This invention employs techniques such as tournament selection and partial matching crossover (PMX) to ensure population diversity and prevent premature algorithm convergence. Tournament selection randomly selects a subset of individuals to compete, preserving superior genes, while partial matching crossover is introduced to generate diverse offspring, thereby maintaining the genetic diversity of the population and enhancing the algorithm's search capability.

[0173] (6) Comprehensive consideration of industrial environmental factors

[0174] This invention fully considers external factors in the industrial environment, such as hardware failure rate, network latency, and communication stability, in its algorithm design. By incorporating these factors into the task allocation model, the task allocation scheme can dynamically adapt to environmental changes, ensuring that the multi-robot system can maintain efficient and stable operation even in the face of uncertainty.

[0175] Next, we will illustrate the above method through a specific implementation case.

[0176] (1) Parameter settings

[0177] Number of robots and valve types: The system is set to have 3 robots and 5 valve types, corresponding to different assembly tasks.

[0178] Genetic algorithm parameters:

[0179] 1) The maximum number of iterations is set to 500 to improve the optimization depth.

[0180] 2) Set the population size to 100 to increase population diversity.

[0181] 3) The crossover probability is 0.95, which further increases the probability of gene exchange.

[0182] 4) The basic mutation probability is 0.05, which maintains population stability through a low mutation rate.

[0183] 5) The number of elites retained is 5 to ensure the preservation of excellent genes.

[0184] 6) The tournament selection size is set to 5 to optimize selection pressure.

[0185] 7) Set the local search interval to once every 20 generations and adjust the optimization frequency.

[0186] 8) The dynamic adjustment mutation rate switch is turned on, and the diversity threshold is set to 0.2 to dynamically adjust the number of elites.

[0187] (2) Valve assembly task data

[0188] Assembly quantity and steps: The five valve types require 3, 4, 5, 4, and 3 assembly quantities respectively, and the number of assembly steps for each type is 5, 6, 7, 5, and 6.

[0189] Assembly time: The theoretical assembly time (in seconds) for each assembly step for each valve type is as follows:

[0190] Type A: 23, 34, 23, 44, 16

[0191] Type B: 13, 24, 23, 34, 42, 32

[0192] Type C: 24, 35, 25, 46, 14, 22, 22

[0193] Type D: 24, 39, 23, 44, 19

[0194] Type E: 33, 35, 26, 45, 19, 34

[0195] Based on the above data, the total number of tasks is calculated as the sum of the products of the assembly quantity of all valve types and the corresponding steps, and a vectorized representation of the task execution time is generated for processing by the genetic algorithm.

[0196] (3) External factors of industrial environment

[0197] Hardware Failure Rate: 1%

[0198] Average network delay (networkDelayMean): 1 second

[0199] Network latency standard deviation (networkDelayStd): 0.5 seconds

[0200] Communication stability probability: 95%

[0201] Task timeout: 300 seconds

[0202] Maximum number of retries (maxRetry): 3

[0203] (4) Specific settings of the genetic algorithm

[0204] Population initialization: Using heuristics and diversity enhancement, random tasks are assigned to the robots in a random order, and 10% random mutation is introduced.

[0205] Fitness calculation: The fitness is based on the maximum total completion time of the robot, taking into account load balance (the ratio of standard deviation to mean).

[0206] Selection method: Tournament selection, with a maximum of 3 participants.

[0207] Crossover method: Partial matching crossover (PMX), applied with a 95% crossover probability.

[0208] Mutation method: Adaptive non-uniform mutation, the mutation probability is dynamically adjusted with the number of generations (exponential decay or linear decay).

[0209] Local search: A hybrid local search is performed every two generations, combining a greedy algorithm and simulated annealing.

[0210] Elite retention: The 6 best individuals are retained. If the population diversity is below 0.2, the number of elites is increased to maintain superior genes.

[0211] (5) Task execution time generation

[0212] Base time: Determined based on valve type and procedure.

[0213] Hardware failure: 1% chance of task time increasing by 50%.

[0214] Network latency: Normal distribution, mean 1 second, standard deviation 0.5 seconds, take non-negative values.

[0215] Unstable communication: There is a 5% probability of increasing communication overhead by a fixed 5 seconds.

[0216] Random fluctuation: The time for each task is multiplied by a random factor of 1 ± 2% to ensure the time is not less than 0.1 seconds. : 3 times

[0217] (6) Parallel computing settings

[0218] Parallel Pool: Attempt to enable a local parallel pool, using all available cores to accelerate fitness calculation and population initialization.

[0219] (7) Results Display and Visualization

[0220] Results output: optimal allocation scheme, total completion time, load balancing degree. For example... Figure 3 As shown, Figure 3 In the diagram, A1-1 represents the first step of the first set of valves for type A valves; B1-1 represents the first step of the first set of valves for type B valves; and the other symbols can be deduced in the same way.

[0221] Visual content such as Figure 4 As shown, this includes: iterative curves of fitness and load balancing; and a graph showing changes in population diversity.

[0222] Task allocation bar chart as follows Figure 5 As shown.

[0223] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0224] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-robot control valve assembly task allocation optimization method based on a hybrid GA algorithm, applied to a multi-robot control valve assembly system, characterized in that, Includes the following steps: Initialize control valve assembly task data: determine the type of control valve to be assembled, the assembly quantity of each type of control valve, and the assembly steps for each type of control valve. Based on the initialized control valve assembly task data, obtain the total number of assembly steps for all types of control valves to be assembled, which is then used as the total number of tasks. Based on preset parameters and the total number of tasks, a hybrid GA algorithm is used to iteratively optimize the task allocation scheme for control valve assembly, specifically including: (1) Initialize the population based on the total number of tasks, where each individual in the population represents a task allocation scheme; (2) Obtain the fitness and load balance of each individual in the current iteration, and record the individual with the best fitness and its corresponding load balance. (3) Determine if each task has timed out and reallocate the timed-out tasks; (4) Based on the preset number of elites to be retained, an elite retention strategy is adopted, and individuals are retained in descending order of fitness, so as to retain multiple elite individuals with the best fitness. (5) Select, crossover, and mutate non-elite individuals to generate a new population; (6) Perform local search optimization on the new population; (7) Merge the elite individuals and the optimized new population to generate a new generation of population; (8) Based on the new generation population, repeat the above steps (2)-(7) until the preset maximum number of iterations is reached, and select the individual with the highest fitness in the record as the best allocation scheme.

2. The multi-robot control valve assembly task allocation optimization method based on a hybrid GA algorithm according to claim 1, characterized in that, Also includes: Based on the optimal allocation scheme and relevant statistical data, draw a fitness history chart, a load balancing history chart, and a task allocation chart.

3. The multi-robot control valve assembly task allocation optimization method based on a hybrid GA algorithm according to claim 1, characterized in that, The total number of tasks is expressed as: Where T represents the total number of assembly steps for all types of control valves to be assembled, i.e., the total number of tasks; V represents the total number of V types of control valves to be assembled; N v S represents the assembly quantity of the control valves of type v to be assembled; v This indicates the assembly steps for the control valve of type v to be assembled.

4. The multi-robot control valve assembly task allocation optimization method based on a hybrid GA algorithm according to claim 1, characterized in that, The initialization of the population based on the total number of tasks specifically includes: Create a zero matrix of size equal to the population size multiplied by the total number of tasks; where the population size is the number of drones. Each individual is initialized using a parallel loop; Randomly arrange the task order and assign the tasks sequentially to the robot with the lowest current load to obtain the initial population matrix; By introducing 10% random mutation, the initial population matrix is ​​optimized to obtain the final initialized population.

5. The multi-robot control valve assembly task allocation optimization method based on a hybrid GA algorithm according to claim 1, characterized in that, The fitness, which is the maximum total completion time of all robot tasks, is expressed as: Where H represents the maximum total completion time of all robot tasks, i.e., fitness; R represents the total number of robots; T represents the total number of tasks; t i Indicates the execution time of task i; δ i,r Let δ represent the task assignment indicator function. If task i is assigned to robot r, then δ i,r It is 1 if it is true, otherwise it is 0.

6. The multi-robot control valve assembly task allocation optimization method based on a hybrid GA algorithm according to claim 1, characterized in that, The load balancing degree is expressed as: Where B represents load balancing degree; T r σ(T) represents the total task time of robot r; r ) represents T r Standard deviation; μ(T) r ) represents T r The mean of the ; Q represents a constant to prevent division by zero; R represents the total number of robots.

7. The multi-robot control valve assembly task allocation optimization method based on a hybrid GA algorithm according to claim 1, characterized in that, Step (3) specifically includes: Obtain the base assembly time for each task; Based on the base assembly time for each task, obtain the task assembly time after considering the hardware failure rate; expressed as: t’ base =t base ×(1+0.5×F) F~Bernoulli(P f ) Among them, t' base This represents the task assembly time after considering hardware failure rate; t base Represents the base assembly time of the task; F represents the hardware fault indication variable; P f This represents the hardware failure rate; Bernoulli represents the Bernoulli distribution. Based on the task assembly time taking into account hardware failure rate, and combined with network latency and communication overhead, the actual execution time of each task is obtained; expressed as: t=max(([t' base ×(1+0.5×F)])+D+C,0.1)×(1+ε) Where t represents the actual execution time of a single task; D represents network latency; C represents communication overhead; and ε represents the fluctuation factor. Determine whether the actual execution time of each task exceeds the preset time. If there are timed-out tasks, trigger the retry mechanism to reassign the timed-out tasks to other robots.

8. The multi-robot control valve assembly task allocation optimization method based on a hybrid GA algorithm according to claim 1, characterized in that, Step (5) specifically includes: The tournament selection method is used to select non-elite individuals; A partial matching crossover method is used to perform crossover operations on selected non-elite individuals to generate offspring; A new population is generated by adjusting the genes that produce offspring through adaptive non-uniform mutation.

9. The multi-robot control valve assembly task allocation optimization method based on a hybrid GA algorithm according to claim 1, characterized in that, Step (6) specifically includes: Based on the total task time of each robot, identify the robot with the heaviest workload; Determine if the robot with the heaviest load is overloaded. If there are overloaded tasks, the task with the longest execution time that the robot is responsible for will be selected and the task will be reassigned to the robot with the lightest load. The simulated annealing algorithm is used to optimize the new task allocation and obtain an optimized new population.

Images