A Robot Assembly Line Balancing Method Based on Hybrid Discrete Particle Swarm Optimization Algorithm

Abstract

The present invention relates to the technical field of assembly line balancing, and particularly to a robot assembly line balancing method based on a hybrid discrete particle swarm algorithm, which includes the following specific steps: 1. Obtain the parameter information of the robot assembly line; 2. Give the calculation formula for the total cost of the robot assembly line; 3. Design a discrete particle swarm optimization algorithm to obtain the optimal process and robot allocation scheme; The present invention updates the process and robot allocation vectors of the particles by using path relinking, multi-segment crossover mechanism, and segment mutation mechanism, and designs a search mechanism specifically for searching the discrete feasible solution space, which is beneficial to improving the search efficiency of the algorithm and the probability of hitting the global optimal solution. The present invention adopts a hybrid discrete particle swarm algorithm, combines the advantages of the exact method dynamic programming and the particle swarm, and has a fast calculation and convergence speed and strong global search ability when solving the robot assembly line allocation problem, and can obtain a high-quality process and robot allocation solution in a short time.

Description

Technical Field

[0001] The present invention relates to the technical field of assembly line balancing, and particularly to a robot assembly line balancing method based on a hybrid discrete particle swarm optimization algorithm. Background Art

[0002] Robots can assist or replace workers in performing repetitive and dangerous tasks, and significantly improve production line efficiency and product quality. Therefore, they are widely installed in assembly lines and used for assembling large commodities. However, the cost of building a robot assembly line is relatively high, and the selection and allocation of robots directly affect the total cost, which is crucial for the competitiveness and long-term development of enterprises.

[0003] Since the assembly line balancing problem belongs to the NP-hard problem, most of the current methods for solving the assembly line balancing problem in patents are meta-heuristic algorithms. For example, the existing patent CN107316107B uses a particle swarm optimization algorithm to solve the warp knitting machine assembly line balancing problem for multi-objective optimization, the patent CN112686474B uses an improved water wave optimization algorithm to solve the parallel assembly line balancing problem, the patent CN113901728A proposes a computer second-class assembly line balancing optimization method based on a migration genetic algorithm, and the patent CN109146136A proposes a method for solving the first-class assembly line balancing problem based on an immune genetic algorithm. The above patents all use meta-heuristic algorithms. Although solutions can be found in a relatively short time, the solutions obtained are often worse than those obtained by exact methods.

[0004] Since the robot assembly line balancing problem also belongs to the NP-hard problem, heuristic and meta-heuristic algorithms have been widely used. However, with the improvement of computer computing power, users tend to look for optimal solutions or near-optimal solutions rather than pursuing extremely short computing times. Hybrid algorithms that combine the advantages of exact methods and meta-heuristic algorithms may have obvious advantages in finding high-quality solutions.

[0005] Among the meta-heuristic algorithms for solving NP-hard problems, the particle swarm optimization algorithm has the advantages of fast convergence speed and high computational efficiency. However, in existing patents, the discrete particle swarm optimization algorithm designed for the assembly line balancing problem has a single function and does not combine exact methods to quickly obtain the optimal solution. Therefore, it is urgent to improve the basic continuous particle swarm optimization algorithm and design a discrete particle swarm optimization algorithm that searches in the discrete feasible space to solve the cost-oriented assembly line balancing problem. Summary of the Invention

[0006] The object of the present invention is to propose a robot assembly line balancing method based on a hybrid discrete particle swarm optimization algorithm for the problems existing in the background art.

[0007] The technical solution of the present invention is a robot assembly line balancing method based on a hybrid discrete particle swarm optimization algorithm, including the following specific steps:

[0008] S1. Obtain the parameter information of the robot assembly line: including the number of processes, the precedence relationship of processes, the number, types of robots, the working time of each robot for each process, and the working cost of each robot for each process;

[0009] S2. Give the calculation formula for the total cost of the robot assembly line:

[0010] The total cost of the robot assembly line, including the purchase cost of the robot and the working cost of the robot for executing the process, is calculated as follows:

[0011]

[0012] Among them, W represents the maximum number of workstations, w represents the workstation number, w = {1, 2,..., W}; R represents the types of robots, r represents the robot number, r = {1, 2,..., R}; T represents the number of processes, t represents the process number, t = {1, 2,..., T}; yrw represents whether robot r is assigned to workstation w, y rw = {0, 1}; x irw represents that process i is assigned to workstation w and executed by robot r, x irw = {0, 1}; PC r represents the purchase cost of robot r; OC ir represents the working cost of process i executed by robot r;

[0013] S3. Design a discrete particle swarm optimization algorithm to obtain the optimal process and robot allocation scheme, including the following specific steps:

[0014] S31. Initialize parameters; set the total number of individuals NP included in the population in the algorithm, the maximum number of iterations gen max , the maximum speed v of the particle max , the individual learning factor C1, the social learning factor C2, the maximum value w of the inertia factor max and the minimum value w min , the probability pc of selecting the individual extreme value and the current global optimal solution, the mutation probability pm, the position update ratio pe; the probability px of selecting path reconnection, the minimum Hamming distance D min ;

[0015] S32. Initialize the population; randomly generate NP particles in the form of permutation number direct coding to form the initial population, and randomly initialize the speed of the particles;

[0016] S33. Calculate the objective function value and fitness value;

[0017] S34. Record the individual extreme value and the current global optimal solution;

[0018] S35. Perform path relinking with probability px and conduct a detailed search near the local optimal solution;

[0019] S36. Select to perform multi - fragment crossover or mutation operation with probability (1 - px);

[0020] S37. Calculate the fitness value of the individual obtained through path relinking, multi - fragment crossover or fragment mutation, and update each individual extreme value P i and the current global optimal solution P g ;

[0021] S38. Update the population, use the updated particles as the new generation population, and increment the iteration count by 1;

[0022] S39. Determine whether the termination condition is met. If it is met, output the optimal solution P g and end. Otherwise, return to step S35.

[0023] Compared with the prior art, the present invention has the following beneficial technical effects:

[0024] 1. The present invention uses path relinking, multi - fragment crossover mechanism, and fragment mutation mechanism to update the process and robot assignment vector of particles, and designs a search mechanism specifically for searching the discrete feasible solution space. This mechanism enables particles to reach any point in the feasible solution space, which is conducive to improving the search efficiency of the algorithm and the probability of hitting the global optimal solution.

[0025] 2. The present invention adopts a hybrid discrete particle swarm algorithm, combining the advantages of the exact method of dynamic programming and particle swarm. When solving the robot assembly line assignment problem, it has a fast calculation and convergence speed, strong global search ability, and can obtain high - quality process and robot assignment solutions in a relatively short time.

[0026] 3. In terms of the optimization objective, the present invention combines the characteristics of robot assembly, and aims to minimize the cost for the assignment of processes and robot types, which is closer to the actual situation, thus reducing the total cost of the assembly line.

[0027] 4. The algorithm proposed by the present invention is not only applicable to the robot assembly line; it is also applicable to the assembly lines of other mechanical equipment. Brief Description of the Drawings

[0028] Figure 1 It is a schematic flow chart of the hybrid discrete particle swarm algorithm in the present invention.

[0029] Figure 2Schematic diagram of the planar structure of the robot assembly line in a specific embodiment of the present invention.

[0030] Figure 3 Schematic diagram of the position vector of particles in the discrete particle swarm algorithm of the present invention.

[0031] Figure 4 Schematic diagram of the path relinking mechanism process in the hybrid discrete particle swarm algorithm of the present invention.

[0032] Figure 5 Schematic diagram of the multi - segment crossover mechanism process in the hybrid discrete particle swarm algorithm of the present invention.

[0033] Figure 6 Schematic diagram of the segment mutation mechanism process in the hybrid discrete particle swarm algorithm of the present invention.

[0034] Figure 7 Precedence relationship diagram of 25 processes in a specific example of the present invention.

[0035] Figure 8 Schematic diagram of the search iteration curve at 300 beats in a specific embodiment of the present invention.

[0036] Figure 9 Schematic diagram of the search iteration curve at 350 beats in a specific embodiment of the present invention.

[0037] Figure 10 Schematic diagram of the search iteration curve at 400 beats in a specific embodiment of the present invention. Specific implementation manners

[0038] A method for balancing a robot assembly line based on a hybrid discrete particle swarm algorithm in the present application first determines the parameters of the robot assembly line balancing problem, and gives the calculation method and formula for the total cost of the assembly line. Then, a discrete particle swarm optimization algorithm is designed to obtain the assembly line balancing scheme. Specifically, according to the precedence relationship of processes, the processes and robots are encoded by using the permutation number direct encoding method, decoded by using dynamic programming, and an initial population is randomly generated. The particles are updated by selecting path relinking, multi - segment crossover, and segment mutation update mechanisms with a certain probability. The individual extreme value and the current global optimal solution are updated, and the search continues until the termination condition is reached. Finally, the obtained optimal solution is decoded to obtain the current optimal solution.

[0039] The following specifically describes a method for balancing a robot assembly line based on a hybrid discrete particle swarm algorithm in this embodiment.

[0040] I. Determine the parameters related to the robot assembly line balancing problem

[0041] In a robotic assembly line, the various components of a product are assembled by assembly robots into a complete product through a logistics transmission system (such as a conveyor belt). The problem of balancing a robotic assembly line aims to reasonably allocate a group of robots and tasks to several workstations, subject to constraints such as precedence relationships, cycle times, and limited resources, so as to achieve goals such as time savings and cost reduction, as Figure 1 shown. The problem of the robotic assembly line with the goal of cost is to obtain the minimum total cost under a given production line beat. The total cost in the present invention includes the robot purchase cost and the working cost of the operation process.

[0042] This patent mainly solves the problem of balancing a robotic assembly line with the following characteristics: only considering the case of assembling a single product on a straight-line assembly line; each task can only be assigned to one workstation and can only be executed by one robot; there are multiple types of robots to choose from, the number of a certain type of robot is more than zero and there is no upper limit; each workstation can only have one robot, and the same type of robot can be assigned to multiple workstations; when a task is executed by different robots, the working time and working cost are different.

[0043] The specific steps for obtaining the relevant parameters of the assembly line balance problem include:

[0044] S1. First, obtain the basic parameters of the robotic assembly line: including the beat, the number of processes, the precedence relationship of the processes, the number, types of robots, the working time of each robot for each process, and the working cost of each robot for each process;

[0045] S2. Give the calculation formula for the total cost of the robotic assembly line:

[0046] The total cost of the robotic assembly line mainly includes the robot purchase cost and the working cost of the robot executing the process. The calculation formula is as follows:

[0047]

[0048] Among them, W represents the maximum number of workstations, w represents the workstation serial number, w = {1, 2,..., W}; R represents the types of robots, r represents the robot serial number, r = {1, 2,..., R}; T represents the number of processes, t represents the process serial number, t = {1, 2,..., T}; y rw represents whether robot r is assigned to workstation w, y rw = {0, 1}; x irw represents that process i is assigned to workstation w and is executed by robot r, x irw = {0, 1}; PC r represents the purchase cost of robot r; OC irDenote the working cost of process i executed by robot r.

[0049] II. Obtain the assembly line balancing scheme that minimizes the cost by using the hybrid discrete particle swarm optimization algorithm

[0050] Design a discrete particle swarm optimization algorithm to obtain the optimal process and robot allocation scheme. The algorithm flowchart is as Figure 2 shown and includes the following steps:

[0051] S1. Initialize parameters: Set the total number of individuals NP in the population in the algorithm, the maximum number of iterations gen max , the maximum velocity v of the particle max , the individual learning factor C1, the social learning factor C2, the maximum value w of the inertia factor max and the minimum value w min , the probability pc of selecting the individual extreme value and the current global optimal solution, the mutation probability pm, the position update ratio pe; the probability px of selecting path relinking, the minimum Hamming distance D min ;

[0052] S2. Initialize the population: Randomly generate NP particles in the form of permutation number direct coding to form the initial population, and randomly initialize the velocity of the particles;

[0053] The position vector of the particle is as Figure 3 shown, where the vector elements correspond to the process numbers;

[0054] The vector of the particle is encoded by direct permutation number and can be expressed as:

[0055] X i =(X1,X2,...,X m ,...,X T ) (2)

[0056] In the formula, i represents the example number, X represents any process, and X i ={1,2,3,...,T}, T represents the number of processes;

[0057] S3. Calculate the objective function value and fitness value: Decode by using the dynamic programming method, calculate the objective function f of each particle, and calculate the fitness value of the particle. The dynamic programming method decomposes the complex task allocation and robot selection process into relatively simple sub-problems, and then solves each sub-problem separately to obtain the optimal solution. This method can solve the optimal solution corresponding to a certain process sequence within a polynomial time.

[0058] Among them, the specific steps of decoding by the dynamic programming method are as follows:

[0059] A. For a given process sequence, create a virtual node C(T + 1) with an empty state that contains no processes or robots.

[0060] B. Starting from the virtual node, recursively calculate backward in sequence from the end of the process sequence using the following recurrence formula:

[0061] C(T + 1) = 0,

[0062]

[0063] where i, j, n represent the process numbers, T represents the number of processes, r represents the robot number, C(i) represents the total cost from process i to the virtual node, PC r represents the purchase cost of robot r, and OC ir represents the working cost of process j performed by robot r .

[0064] C. According to the above recursive process, node i has multiple states, each containing different processes and robot types. Therefore, solving the problem of the minimum total cost for a given process is transformed into solving the problem of the minimum path between the start and end nodes. When the loop ends, the optimal solution is obtained through backtracking.

[0065] S4. Record the individual extreme value and the current global optimal solution: Record the initialized individual as the individual extreme value of the particle. The individual extreme value of particle i is denoted as P i , and record the individual with the minimum objective function value as the current global optimal solution P g ;

[0066] S5. Perform path relinking with probability px: In the path relinking mechanism, there are two parent particles. Parent one is the current solution, serving as the initial solution in path relinking, and parent two is the global optimal solution, serving as the guiding solution in path relinking, to construct a path connecting the two solutions. As Figure 4 shown, it specifically includes the following steps:

[0067] A. Traverse the initial solution from front to back, compare the processes at the corresponding positions of the initial solution and the guiding solution. If they are different, find the position of the current process in the guiding solution and exchange them. If they are the same, continue to traverse the next position of the initial solution;

[0068] B. After the exchange, determine whether the new solution is feasible. If it is feasible, calculate its fitness value. If the fitness value of the new solution is better than that of the initial solution, replace the initial solution in the population;

[0069] C. Repeat steps A and B until traversing to the last process of the initial solution. At this time, the initial solution becomes the guiding solution through exchanges.

[0070] S6. Select to perform multi - fragment crossover or mutation operations with a probability of (1 - px). The basis for crossover and mutation is as follows: If the Hamming distance between the current individual and the optimal individual is greater than or equal to 1 / 2 of the chromosome length, i.e., D min , it is considered that the current individual can still evolve towards the optimal individual, and perform multi - fragment crossover operations between the current individual and the optimal individual. If the Hamming distance between the current individual and the optimal individual is less than 1 / 2 of the chromosome length, it is considered that the distance between the current individual and the optimal value is too close, which is not conducive to population diversity. At this time, perform fragment mutation on the current individual.

[0071] S61. Among them, the steps of updating particles by the multi - fragment crossover mechanism are as follows, which can be referred to Figure 5 :

[0072] A. Update the velocity of the particle. Use a linear inertia factor, the inertia factor ω gen As the search generation decreases from the maximum value to the minimum value, improve the global search ability in the early stage and the local search ability in the later stage of the algorithm;

[0073] Update the particle velocity using the following formula:

[0074] V i = ω gen ×V i - C1×rand()×(p i - X i ) + C2×rand()×(p g - X i ) (4)

[0075] Among them, V i represents the velocity vector of particle i, C1 represents the individual learning factor of the particle, C2 represents the social learning factor of the particle, rand() represents a randomly generated number between 0 and 1, and ω gen represents the inertia factor, which adopts a linear decreasing weight strategy, and the calculation formula is as follows:

[0076] ω gen = ω max - (gen - 1)(ω max - ω min ) / (gen max - 1) (5)

[0077] Among them, ω gen and ω min respectively represent the maximum and minimum values of the inertia factor, gen represents the number of search times, and gen max represents the maximum number of search times;

[0078] B. Normalize the velocity particle V i (t) to obtain ||Vi (t) ||;

[0079] Particle velocity V i (t) The normalization process is calculated as follows:

[0080]

[0081] Among them, T represents the total number of processes, ||V i (t) || represents the update probability of the elements in the velocity vector;

[0082] C. Select M elements with higher update probability from the parent two X R (t) and obtain M or (M + 1) segments;

[0083] The selection process of M elements with higher probability is as follows:

[0084]

[0085] In the formula, p e represents the position update ratio, and T represents the total number of processes;

[0086] D. Directly copy the parent two X R (t) to generate a child X i (t + 1);

[0087] E. Mark the elements in the parent one X i (t) that are different from the parent two X R (t) and record the number of elements as X;

[0088] F. Select a segment from the child X i (t + 1), record the number of elements as Y, and mark the same elements in the parent one X i (t), and record the number of elements as y;

[0089] G. Adjust the y marked elements in the parent one X i (t) in order from left to right to generate a new segment, replace the first y elements in the child X i (t + 1) segment, and cancel the mark in the parent one X i (t);

[0090] H. If y < Y, select the first (Y - y) elements from the X elements, cancel the mark, and replace the remaining elements in the child segment, and at the same time make X = X - (Y - y);

[0091] I. Determine whether all segments have been adjusted. If so, end. Otherwise, go to step D;

[0092] S62. Among them, the fragment mutation mechanism can be referred to Figure 6 , and the specific steps are as follows:

[0093] A. Randomly select a fragment from the current individual;

[0094] B. Re - sort according to the precedence relationship of this fragment to obtain a new feasible fragment;

[0095] C. Replace the original fragment with this new fragment to generate a new feasible individual.

[0096] S7. Update the individual extreme value and the current global optimal solution: Calculate the objective function and fitness value of the individuals obtained through path relinking, multi - fragment crossover, and fragment mutation, and compare and update each individual extreme value P i and the current global optimal solution P g ;

[0097] S8. Update the population: Use the updated particles as the new generation population and increment the iteration count by 1;

[0098] S9. Determine whether the termination condition is met. If it is met, output the optimal solution P g and end. Otherwise, return to step S35.

[0099] Example 1

[0100] Experimental analysis is carried out using the actual data of the robot assembly line in the assembly workshop of a certain automobile company as follows:

[0101] 1. Parameter setting

[0102] There are 25 processes for this assembled product, and its precedence relationship of processes is as Figure 7 shown. According to the characteristics of the assembly processes, 6 different types of robots are provided. Among them, the purchase cost of each robot is shown in Table 1, the working time of each robot to complete each process is shown in Table 2, and the working cost per unit time of each robot to execute each process is shown in Table 3. To fully verify the performance of the algorithm, the cycle times are set to 300, 350, and 400 respectively, with a total of 3 groups of cases.

[0103] Among them, the parameters of the hybrid discrete particle swarm algorithm are shown in Table 4, and the selected parameters are all common parameters. To ensure the accuracy of the comparison, the parameters selected for the basic particle swarm algorithm are shown in Table 5, and the key parameters are the same as those of the hybrid discrete particle swarm algorithm. The parameters selected for the genetic algorithm are shown in Table 6, and the key parameters are the same as those of the particle swarm algorithm.

[0104] Table 1 Purchase cost of each robot

[0105]

[0106] Table 2 Working time of each robot for each process

[0107]

[0108]

[0109] Table 3 Working cost per unit time of each robot for each process

[0110]

[0111] Table 4 Parameters of hybrid discrete particle swarm optimization algorithm

[0112]

[0113] Table 5 Parameters of basic particle swarm optimization algorithm

[0114]

[0115] Table 6 Parameters of genetic algorithm

[0116]

[0117] 2. Analysis of experimental results

[0118] First, the hybrid discrete particle swarm optimization algorithm is used to solve the problem, and the optimization results are compared with those of the genetic algorithm and the basic particle swarm optimization algorithm to verify the superiority of the discrete particle swarm optimization algorithm. The algorithm is implemented in C++ programming using Visual Studio 2013 software, and the model is solved 10 times respectively.

[0119] The optimal results and average values of 25 processes in the example under 3 beats are shown in Table 7. The iteration curves of the three algorithms are shown in Figure 8 .

[0120] Table 7 Comparison table of results of genetic algorithm, basic particle swarm optimization algorithm and hybrid discrete particle swarm optimization algorithm

[0121]

[0122] From the comparison in Table 7, it can be obtained that when optimizing for the goal of minimizing the total cost, the quality of the solutions obtained by the hybrid discrete particle swarm optimization algorithm is significantly better than that of the genetic algorithm and the basic particle swarm optimization algorithm. Figure 8 The comparison results of the iteration curves in also show that the discrete particle swarm optimization has a fast convergence speed. Therefore, the hybrid discrete particle swarm optimization algorithm proposed in the present invention has strong search ability and convergence performance, can obtain a high-quality allocation scheme, and is effectively applied to the robot assembly line balancing problem.

Claims

1. A robot assembly line balancing method based on a hybrid discrete particle swarm algorithm, characterized in that It includes the following specific steps: S1. Obtain the parameter information of the robot assembly line: including the number of processes, the precedence relationship of processes, the number and types of robots, the working time of each robot for each process, and the working cost of each robot for each process; S2. Give the calculation formula for the total cost of the robot assembly line: The total cost of the robot assembly line, including the purchase cost of the robot and the working cost of the robot for performing processes, is as follows: Among them, represents the maximum number of workstations, represents the serial number of the workstation, ; represents the type of robot, represents the serial number of the robot, ; represents the number of processes, represents the serial number of the process, ; represents whether the robot is assigned to the workstation ; ; represents the process assigned to the work and is executed by the robot ; ;PC r represents the purchase cost of the machine ; OC ir represents the working cost of the process executed by the robot ; S3. Design a discrete particle swarm optimization algorithm to obtain the optimal process and robot allocation scheme, including the following specific steps: S31. Initialize the parameters; S32. Initialize the population; randomly generate particles in the way of direct permutation encoding to form an initial population, and randomly initialize the velocities of the particles; S33. Calculate the objective function value and fitness value: Decode using the dynamic programming method and calculate the objective function of each particle , and calculate the fitness value of the particle; The dynamic programming method decomposes the complex task assignment and robot selection process into relatively simple sub-problems, and then solves each sub-problem separately to obtain the optimal solution; Among them, the specific steps of decoding by the dynamic programming method are as follows: A. For a given process sequence, create a virtual node , whose state is empty and does not contain any processes or robots; B. Starting from the virtual node, recursively deduce from the end of the process sequence forward in turn according to the recurrence formula, and the recurrence formula is shown as follows: , Among them, represents the serial number of the process, represents the number of processes, represents the robot serial number, and C(i) represents the total cost from the process to the virtual node, PC r represents the purchase cost of the robot OC ir represents the process The working cost executed by the robot ; C. According to the above recurrence process, the node has multiple states, and each state contains different processes and robot types; therefore, the problem of solving the minimum total cost for a given process is transformed into the problem of finding the minimum path between the start and end nodes; when the loop ends, the optimal solution is obtained by backtracking; S34. Record the individual extreme value and the current global optimal solution; S35. Perform path reconnection with probability px and conduct a detailed search near the local optimal solution; S36. With a probability select to perform multi - fragment crossover or fragment mutation operations; S37. Calculate the fitness value of the individuals obtained through path relinking, multi-fragment crossover, or fragment mutation, and compare and update the extreme value of each individual and the current global optimal solution ; S38. Update the population, use the updated particles as the new generation population, and increment the iteration count by 1; S39. Determine whether the termination condition is satisfied. If it is satisfied, output the optimal solution and end. Otherwise, return to step S35.

2. The robotic assembly line balancing method based on a hybrid discrete particle swarm optimization algorithm according to claim 1, wherein In S34, the individual initialized and formed is recorded as the individual extreme value of the particle, and the particle 's individual extreme value is represented as , and the individual with the minimum objective function value is recorded as the current global optimal solution .

3. A robot assembly line balancing method based on a hybrid discrete particle swarm optimization algorithm according to claim 1, characterized in that In the path reconnection mechanism in S35, it includes two parent particles. Parent one is the current solution, serving as the initial solution in path reconnection, and parent two is the global optimal solution, serving as the guiding solution in path reconnection, to construct a path connecting the two solutions; specifically including the following steps: A. Traverse the initial solution from front to back, compare the processes at the corresponding positions of the initial solution and the guiding solution. If they are different, find the position of the current process in the guiding solution and exchange it. If they are the same, continue to traverse the next position of the initial solution; B. Judge whether the new solution after the exchange is feasible. If it is feasible, calculate its fitness value. If the fitness value of the new solution is better than that of the initial solution, replace the initial solution in the population; C. Repeat steps A and B until traversing to the last process of the initial solution. At this time, the initial solution becomes the guiding solution through exchange.

4. A robot assembly line balancing method based on a hybrid discrete particle swarm algorithm according to claim 1, characterized in that The basis for crossover or mutation in S36 is that the Hamming distance between the current individual and the optimal individual is greater than or equal to 1 / 2 of the chromosome length, i.e., D min , perform a multi-fragment crossover operation between the current individual and the optimal individual; The Hamming distance between the current individual and the optimal individual is less than 1 / 2 of the chromosome length. At this time, perform segment mutation on the current individual.

5. A robot assembly line balancing method based on a hybrid discrete particle swarm optimization algorithm according to claim 4, wherein, The specific steps of the multi-segment crossover mechanism in S36 are as follows: A. Update the velocity of the particles, using a linear inertia factor, the inertia factor ω gen As the search generation decreases from the maximum value to the minimum value, improve the global search ability in the early stage and the local search ability in the later stage of the algorithm; B. Normalized velocity particle V i (t) to obtain ||V i (t)||; C. From parent two Select the elements with a higher update probability And obtain Or Fragments; D. Directly copy Parent Two Generate a child ; E. In Parent One Mark the elements different from Parent Two and record the number of elements as ; F. From the offspring Select a segment, and record the number of elements as , and mark the elements with the same label in Parent 1 , and record the number of elements as ; G. Take Parent One The marked elements are adjusted in the order from left to right to generate a new segment, replacing the first several elements in the segment of the child, and the markings in Parent One are cancelled; H. If , then select the first elements from elements, unmark them, and replace the remaining elements in the child segment. At the same time, let ; I. Judge whether all segments have been adjusted. If so, end. Otherwise, return to step D.

6. A robot assembly line balancing method based on a hybrid discrete particle swarm optimization algorithm according to claim 4, characterized in that The specific steps of the segment mutation mechanism in S36 are as follows: A. Randomly select a segment from the current individual; B. Reorder according to the precedence relationship of this segment to obtain a new feasible segment; C. Replace the original segment with this new segment to generate a new feasible individual.

Images