Multi-vehicle-type mixed-line final assembly process optimization method based on load balancing

Abstract

The invention discloses a multi-vehicle-type mixed-line final assembly process optimization method based on load balancing, and belongs to the technical field of automobile manufacturing. Aiming at a multi-variety mixed assembly line balance problem, the method optimizes process distribution through a genetic algorithm, and comprises the following steps: inputting production vehicle type, process, unit and operation time data, and initializing genetic algorithm parameters; generating an initial population meeting a process dependency relationship; a filial generation population is generated through selection, crossover and mutation operation, a fitness function quantifies multi-vehicle-type global balance performance based on weighted load mean square errors, and an elitist retention and roulette selection strategy is adopted; and correcting individuals violating process constraints, and outputting an optimal process allocation scheme after iterative evolution. According to the method, comprehensive load balancing can be realized on all target vehicle models, and the adaptability and efficiency of a production line are improved.

Description

Technical Field

[0001] This invention belongs to the field of automotive manufacturing technology, specifically relating to a method for process optimization and balancing of mixed-line automotive assembly production lines, and in particular a method for optimizing the final assembly process of multi-model mixed-line assembly based on load balancing and intelligent optimization algorithms. Background Technology

[0002] With increasingly personalized consumer demands, the modern automotive manufacturing industry has largely shifted from traditional large-scale production of a single model to a mixed-model assembly line (MMALBP) production model, where multiple different models are produced on the same assembly line. This shift has transformed the production line balancing problem from the relatively simple single-model balancing problem to a more complex and dynamic mixed-model assembly line balancing problem. Unlike the traditional single-model production line balancing problem, the mixed-model assembly line balancing problem is far more complex. The core objective of MMALBP is no longer to find the optimal balance solution for a single, static production process, but rather to design an adaptive and robust general process allocation scheme for a dynamic and changing production environment.

[0003] Existing solutions to this problem have significant limitations. First, directly applying optimization methods specific to a single vehicle model has a fundamental adaptability flaw. Due to significant differences in process time, assembly sequence, and resource requirements across different vehicle models, traditional optimization schemes focused solely on a single model can lead to severe load imbalances during model switching, failing to meet the basic requirement of flexible manufacturing for cross-model versatility. Second, while exact algorithms, such as Mixed Integer Linear Programming (MILP), can theoretically find the optimal solution, their computational complexity increases exponentially with the number of processes, workstations, and vehicle models. For large-scale, complex real-world scenarios, they often lack engineering practicality due to computational infeasibility.

[0004] Therefore, this invention designs a multi-model mixed-line final assembly process optimization method based on load balancing to ensure that the load of the production line can be maintained at a near-optimal balanced state when producing any target model. Summary of the Invention

[0005] To address the above technical problems, this invention proposes a method for optimizing the multi-model mixed-line final assembly process based on load balancing, comprising the following steps: Step 1: Input the basic data for mixed-line production, including the set of production models M, the set of production processes J, the set of production units K, and the standard operating time of each model i on each process j. Set the operating parameters of the genetic algorithm, including population size N and crossover rate. Variation rate The maximum number of generations (Gen) is determined, and the running parameters are initialized.

[0006] Step 2: Generate an initial population according to the encoding scheme, wherein the encoding scheme represents a process allocation scheme as a string of length [length missing]. The initial population consists of individual sequences, where each position in the sequence corresponds to a process step, and its gene value represents the production unit number to which that process step is assigned. A genetic algorithm is used to randomly generate N individual sequences that satisfy the process step dependency relationships, forming the initial population. ; Step 3: For each individual, calculate the weighted mean square error of the load across all vehicle models based on the process allocation scheme it represents. Based on this, its fitness value (Fitness) is calculated using a negative exponential function; Step 4: For the current population The selection, crossover, and mutation operations are performed sequentially to generate the offspring population. ; Step 5: Check whether the offspring individuals violate the process dependency constraints. If they do, correct them through the feasibility guarantee mechanism to ensure that all individuals are feasible solutions. Step 6: Transfer the corrected offspring population As the next generation population P t+1 Determine whether the current generation t has reached the set maximum value Gen. If not, return to step 3; if it has, terminate the algorithm and output the process allocation scheme corresponding to the individual with the highest fitness value in the entire evolution process as the final optimization result.

[0007] Furthermore, the encoding scheme specifically includes: the individual sequence is represented as follows: ... ,in , Index i represents the process number, gene value This indicates the production unit number to which the process is assigned.

[0008] Furthermore, the following constraints must be satisfied when generating the initial population in step 2 and when correcting infeasible solutions in step 5: Process dependency constraint: For any vehicle model If the process Must be in the process If completed previously, then the following conditions must be met. ,in Let be the completion time of process j. It is the process of model i Standard operating time; Process unique assignment constraint: Each process must and can only be assigned to one production unit, that is, for any process... ,have ,in A binary variable representing a process. Whether it has been assigned to production unit k; Multi-model consistency constraint: All models use the same process-unit mapping structure, that is, for any model ,have .

[0009] Furthermore, the calculation steps for the fitness value (Fitness) mentioned in step 3 are as follows: For the current individual sequence, iterate through each vehicle type i∈M; ② Calculate the total working hours of vehicle model i in production unit k. ; ③ Calculate the average total working hours of vehicle model i across all production units. ; ④ Calculate the load mean square error of vehicle model i ; ⑤ Calculate the weighted global comprehensive evaluation index ,in Let be the weighting coefficient for vehicle model i, and ; ⑥ Calculate the fitness function , where α is a constant.

[0010] Wherein, the weighting coefficient It can be set according to changes in vehicle model business needs and market demand.

[0011] Furthermore, the selection operation in step 4 employs a strategy combining elite retention and roulette wheel betting, specifically including: Elite selection: The top 10% of individuals in the current population based on their fitness values ​​are directly copied to the next generation of the population; Roulette wheel selection: From the remaining population after removing elite individuals, the selection probability is calculated based on the proportion of individual fitness value to total fitness, and individuals that account for 90% of the population size are selected by roulette wheel selection.

[0012] Furthermore, the specific process of the crossover operation described in step 4 is as follows: For each production unit k, calculate its comprehensive load standard deviation across all vehicle models in both parent individuals. ; Compare the combined load standard deviation of the same production unit k in the two parent generations. Inheritance in offspring has a smaller chance of success. The process combination scheme of the unit in the parent individual of the value.

[0013] The formula for calculating the comprehensive load standard deviation is as follows: , It is the total man-hours of model i in production unit k. For unit Weighted average working hours when carrying all different vehicle models.

[0014] Furthermore, the specific process of the mutation operation described in step 4 is to perform an exchange mutation operation on the offspring individuals generated after crossover at a preset mutation rate, and randomly exchange the production units allocated by the two processes.

[0015] Furthermore, after the crossover and mutation operations in step 4, the process integrity conflict detection and repair is also included: detecting whether there are duplicate or missing processes in the offspring individuals, and ensuring that each process is uniquely assigned by reallocation.

[0016] Furthermore, the feasibility guarantee mechanism described in step 5, for cross-workstation conflicts, uses a global relocation strategy to migrate the process causing the conflict to a new workstation that meets timing constraints and has a better load.

[0017] Furthermore, the global relocation strategy includes the following steps: a. Locate the conflicting processes and the workstations containing their preceding processes; b. Generate a list of all candidate workstations that meet the timing constraints; c. Assess the current load of the candidate workstations; d. Select the candidate workstation with the lowest load as the target; e. Move the conflicting process to the target workstation.

[0018] Compared with the prior art, the beneficial effects of the present invention are: The universal process allocation scheme proposed in this invention can simultaneously adapt to the mixed-line production needs of multiple different vehicle models, effectively overcoming the load imbalance problem caused by traditional single-vehicle optimization schemes when switching vehicle models, and significantly improving the flexibility and adaptability of the production line.

[0019] By employing a genetic algorithm for optimization and designing an efficient encoding scheme and fitness function, this invention avoids the drawbacks of mixed-integer linear programming and other methods, which suffer from high computational complexity and difficulty in solving large-scale problems.

[0020] By constructing a fitness function based on the weighted mean square error of the load, this invention can accurately quantify and optimize the global load balance under mixed production lines for multiple vehicle models.

[0021] This invention introduces a complete constraint system that includes process dependency constraints, unique allocation constraints, and multi-model consistency constraints, and designs a hierarchical repair feasibility guarantee mechanism to ensure that the optimization process is always carried out in a feasible solution space. The final output process allocation scheme strictly conforms to the actual production process requirements and has high executability.

[0022] By transforming the complex multi-objective load balancing problem into a well-structured single-objective optimization problem, and by using a fitness function with negative exponential transformation to guide the search direction, the genetic algorithm can quickly and effectively search for an approximate optimal solution with extremely high load balancing. Attached Figure Description

[0023] Figure 1 Flowchart of optimization method for multi-model mixed-line final assembly process; Figure 2 The design scheme is for vehicle model A; Figure 3 This is a design scheme for vehicle model B; Figure 4 The design scheme is for vehicle model C; Figure 5 This invention optimizes the solution for all target vehicle models simultaneously (each model has the same weight); Figure 6 Comparison of MSE of different solutions for mixed-line production; Figure 7 Bar charts showing the load performance of different solutions for each vehicle model; Figure 8 This invention optimizes the scheme for all target vehicle models simultaneously (each model has a different weight). Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0025] Example The flowchart of the multi-model mixed-line final assembly process optimization method based on load balancing of the present invention is as follows: Figure 1 As shown, it includes the following steps: Step 1: Input the basic data for mixed-line production, including the set of production models M, the set of production processes J, the set of production units K, and the standard operating time of each model i on each process j. Set the parameters for the genetic algorithm, including population size N and crossover rate. Variation rate The maximum number of generations (Gen) is determined, and these runtime parameters are initialized.

[0026] The specific parameter settings are related to the types of vehicle models (M), the total number of production processes (J), and the total number of production units (K). In a scenario containing 3 products, 24 processes, and 12 workstations, the population size (N) can be set to 100, the maximum number of generations (T) to 200, and the crossover rate... Set to 0.8, mutation rate Set to 0.01. The algorithm parameters should be adjusted appropriately depending on the number of vehicle models, processes, and workstations.

[0027] Simultaneously, a coding scheme is defined for the problem: a feasible process allocation scheme is encoded as an individual sequence of length J (i.e., the total number of processes). Each position in the sequence corresponds to a process, and the gene value at that position represents the production unit number (i.e., workshop number or workstation number) to which this process is allocated.

[0028] Encoding Structure: The index of each one-dimensional integer array represents a unique process number (e.g., from 0 to J-1). The value at that index represents the production unit number, or workshop number, to which that process is assigned (e.g., from 0 to K-1). This direct and compact encoding method has high computational efficiency and can seamlessly integrate with standard operators such as crossover and mutation in genetic algorithms, laying the foundation for subsequent global search.

[0029] Table 1. Indices and values ​​at the index positions of the one-dimensional integer arrays formed in practical applications (Model A) Table 2. Indices and values ​​at the index positions of the one-dimensional integer arrays formed in practical applications (Model B) Table 3. Indices and values ​​at the index positions of the one-dimensional integer arrays formed in practical applications (Model C) Table 4 shows the indices and values ​​at the index positions of the one-dimensional integer arrays formed in practical applications (models A, B, and C). Individual definition: A complete one-dimensional integer array of length J (e.g., [0,6,8,...,11]) constitutes a complete process allocation scheme and is the basic unit for genetic algorithms to operate and optimize, i.e., an individual.

[0030] According to the above encoding scheme, an individual sequence can be represented as: ... ,in , Each process in the individual sequence is represented by an index i, and the production unit assigned to the process is determined by the value of its corresponding index. To express.

[0031] The individual sequence defines only a general physical process-production unit allocation relationship, without including any specific vehicle model's work hour information. This design effectively separates the allocation scheme from its performance across different vehicle models, allowing its performance across all vehicle models to be evaluated using a single genotype.

[0032] It is important to note that a valid process allocation scheme represented by an individual sequence must satisfy the "process dependency constraint" described in the background technology and subsequent steps, that is, the allocation order of processes in the sequence cannot violate their inherent process sequence relationship.

[0033] Step 2: Use a genetic algorithm to randomly generate N individual sequences that satisfy the sequential dependency relationship of the processes, and set them as the initial population. .

[0034] To ensure the feasibility of the process allocation scheme generated by the algorithm in actual production, the present invention introduces the following constraints: ① Process Dependency Constraints During assembly, some processes have a strict sequential relationship. This constraint ensures that the generated process allocation plan strictly conforms to the product assembly logic, preventing situations where subsequent processes start before the preceding processes are completed.

[0035] If any car model process Must be in the process Previously completed, therefore, this invention defines Let be the completion time of process j. Then the constraint condition can be expressed as: in, It is the process of model i Standard operating time.

[0036] ② Unique process allocation constraint Each process must be assigned to one and only one production unit. Let binary variables... Let indicate whether process j is assigned to production unit k (1 for yes, 0 for no), then: This constraint ensures that all processes are allocated reasonably, without omissions or duplications.

[0037] ③ Consistency constraints for multiple vehicle models To enable mixed-model assembly on the same production line, all models must use the same process-unit mapping structure, i.e.: Step 3: Addressing the complex characteristics of multi-model mixed-line production, this invention designs a fitness function (Fitness) to quantify the overall load balancing of mixed-line production. This function is used to accurately evaluate the global load balancing performance of each individual (i.e., the process allocation scheme) under different vehicle model combinations. The design steps of this fitness function are as follows: ① For the allocation scheme represented by the current individual sequence, traverse each vehicle type i∈M.

[0038] ②According to vehicle model Dedicated working hour data and binary decision variables Calculate the total working hours of this model in production unit k. for: That is, when production unit is 1 (k=1), the total working hours of model i in production unit 1 are When there are 2 production units (k=2), the total working hours for model i in production unit 2 are: .

[0039] ③ Calculate the average total working hours for this model across all production units. for: ④ Calculate the load mean square error (MSE) corresponding to this vehicle model, denoted as This value quantifies the load balancing degree of the individual sequence when producing model i: ⑤ In actual production, if the production line for a core model is unbalanced, the negative impact is far greater than that of small-batch or peripheral models. To reflect this difference in business priority, this invention assigns a weighting coefficient to each model i. ,in The weights can be set based on the annual production volume or market demand of each vehicle model. It's important to note here that: "Set a weight coefficient for each vehicle model i". The goal is to build a parameterizable optimization framework that can adapt to changing business needs. In real-world production environments, the market value, production priority, or data reliability of different vehicle models (A, B, C) may vary significantly. This is addressed by introducing weighting coefficients. The framework of this invention allows users to adjust the objective function according to specific business objectives (e.g., focusing on optimizing the performance of a high-value vehicle model A). This ensures the adaptability of this invention in different application scenarios. Figure 5 The statement that "all vehicle models have the same weight" is a specific example under this general framework. It represents a "no-preference" baseline setting across all vehicle models (i.e., each of the three models A, B, and C has a weight of 1 / 3), aiming to fairly demonstrate the overall performance of the scheme when balancing all objectives. Therefore, the latter is a specific application example of the former under specific parameter configurations, and the two complement each other. Figure 8 The figure shows the optimization results under different weight parameters.

[0040] Calculate the weighted global comprehensive evaluation index : ⑥ Employ a more robust negative exponential function to integrate the global comprehensive evaluation indicators. Perform the transformation. Fitness function. as follows: like Figure 6 and Figure 7 As shown, compared with the solutions specifically designed for each vehicle model, the overall load balancing of the present invention under mixed-line production (…) There has been a significant improvement, and the load performance is more balanced across different vehicle models. It is worth noting that: Figure 6 Finally, the total error of each solution across all vehicle models is calculated to compare and evaluate their "overall performance." An unweighted scalarization of the MSE (Mean Economic Estimate) for all models provides a fair benchmark. This demonstrates that solutions optimized only for a single vehicle model perform poorly on other models, thus showcasing the superior overall performance of this invention by optimizing for multiple vehicle models simultaneously. The "total error" is a fixed scale set for comparison, while the weighting coefficients... These are the parameters that guide the algorithm to find specific trade-off points during the optimization process. Figure 6 Yes Figures 2-5 Data aggregation Figure 6 For example, the MSE values ​​for each column in the data are 41, 2289, 1344, and 335 for model A. Figure 7 The four pillars of model A in the middle section. Figure 7 This is a cross-comparison of the MSE performance of four optimization strategies across three target vehicle models (A, B, C). It more intuitively demonstrates the limitations of optimizing for a single vehicle model; while achieving "optimality" for its specific objective, performance degrades (MSE spikes) when applied to other objectives. In contrast, the "all-vehicle-specific solution" exhibits robustness across all three models. Although it doesn't reach ultimate optimality on any single model, it achieves acceptable MSE across all objectives by sacrificing performance on a single point, demonstrating its superiority in multi-objective comprehensive optimization scenarios.

[0041] Step 4: Apply genetic operators such as selection, crossover, and mutation to the current population. Perform the operation to generate a offspring population. : Selection Operation: In each generation of the genetic algorithm, a selection strategy combining elite retention and roulette wheel selection is employed. Elite Selection: Individuals in the current population are sorted from highest to lowest fitness value. The top 10% of individuals are directly copied to the next generation without any modification, prioritizing the retention of individuals with the highest overall load balance to ensure the transmission of globally optimal information. Roulette Wheel Selection: To replenish the remaining population size (i.e., 90% of the total population size), from the remaining population after removing elite individuals, a roulette wheel selection method is used to select from the remaining 90% of individuals based on their fitness value. Individuals with higher fitness have a greater probability of being selected.

[0042] Because in the roulette wheel selection method, the probability of each individual being selected is directly proportional to its fitness value relative to the total fitness value of the population. An individual with higher fitness occupies a larger sector on the roulette wheel, and therefore has a higher probability of being selected by the random number pointer. The specific operation process is as follows: ① Calculate total fitness: Calculate the sum of the fitness of all individuals in the current population. ; ② Calculate the selection probability: For each individual, calculate the probability of being selected. ; ③ Calculate the cumulative probability: Sum the selection probabilities of all individuals to obtain a cumulative probability distribution, where each individual corresponds to a cumulative probability value Q in the interval [0,1]. i The cumulative probability distribution is an increasing sequence, with the cumulative probability of the last individual being 1. ④ Generate random number: Generate a random number in the interval [0,1]; ⑤ Identify the individual: Determine which individual's cumulative probability interval the random number falls into; that individual is then selected. Specifically, find the first individual that satisfies Q. i Individual i with a value ≥ r is selected. ⑥ Repeat steps ④ and ⑤ until the number of selected individuals reaches the remaining quota required for the next generation population (i.e., 90% of the population size).

[0043] Crossover: Pair selected parent individuals with a preset mutation rate. Perform a crossover operation. This operator calculates the load standard deviation. To evaluate the overall load performance of the same production unit in two parent generations across all vehicle models, and to prioritize the inheritance of the unit with higher overall load balance and its internal process combinations into the child generation.

[0044] During the crossover operation phase, this embodiment designs a crossover strategy based on the comprehensive load standard deviation to maintain the overall balance among multiple models during the genetic recombination process, taking into account the characteristics of multi-model mixed-line production.

[0045] Because different weights need to be processed For this vehicle model, the "comprehensive load standard deviation" should be the weighted standard deviation, and its calculation steps are as follows: For each production unit Computing unit Weighted average load when carrying all different vehicle models ,Right now: Then in the parent generation and The calculations are performed separately for each production unit k across all vehicle models. The formula for calculating the comprehensive load standard deviation is as follows: in, It is the total man-hours of model i in production unit k. For production units Weighted average working hours when carrying all different vehicle models.

[0046] Calculate the parent generation separately and The overall load standard deviation of each production unit is calculated and compared one by one. A load standard deviation smaller than 1 indicates that it exhibits higher balance across multiple vehicle models. Production units with smaller load standard deviations and their internal process combination schemes are inherited into their offspring.

[0047] Mutation: For the offspring individuals generated after crossover, a preset mutation rate is applied. Perform a swap mutation operation to randomly swap the production units assigned to the two processes.

[0048] To ensure that the offspring generated by the algorithm meet actual production requirements, this invention incorporates conflict detection and repair mechanisms after both crossover and mutation operations. The specific process is as follows: First, after crossover, the exchange of parent gene segments may lead to duplicate or missing steps in the offspring's sequence. Therefore, step integrity conflict detection and repair are necessary. Detection Algorithm: Traverse the standard process list, using a hash table to create a counter for each process and initializing it to 0. Check each production unit and its assigned process in the newly generated child process. Whenever a process is encountered in the child process, increment its corresponding count by 1. After traversal, scan the counter; any process with a count of 0 is identified as a missing process; any process with a count greater than 1 is identified as a duplicate process.

[0049] Repair algorithm: After identifying all missing and duplicate processes in the offspring scheme, all missing processes are randomly reassigned to the positions of processes that were vacated due to duplication, thereby ensuring that each process is uniquely assigned.

[0050] After completing the above repairs, perform the mutation operation. The mutation operation is achieved by randomly swapping the production unit numbers of the two processes. This operation may disrupt the sequential dependencies between processes. Therefore, process dependency conflict detection must be performed on the mutated offspring individuals. Detection Algorithm: Based on a product process list, a list of "process dependencies" containing the sequence of all processes is constructed. Each dependency pair (preceding process P, subsequent process S) is iterated through. For each (P, S) pair, the production unit number K assigned to it is obtained from the child individual code array. pre and K suc The condition for determining inter-station conflicts is: if K pre >K suc If a conflict is detected, it is determined that a conflict exists. This indicates that the subsequent process S has been assigned to a production unit with a lower sequence number (i.e., earlier on the production line), while its preceding process P is in a unit with a higher sequence number (later on the production line), violating the process logic. It should be noted that the coding scheme used in this invention only defines the affiliation relationship between processes and production units, but does not define the execution order of processes within the same unit. Therefore, conflict detection only applies to inter-station conflicts, not intra-station conflicts.

[0051] If a violation of process dependency constraints is detected, a built-in feasibility guarantee mechanism is activated for correction. This mechanism addresses conflicts between workstations (production units) by employing a global relocation strategy. The core of this strategy is to find a new location for the violating downstream process that satisfies both process timing requirements and minimizes disruption to production line balance. The selection criteria are: ① Timing feasibility, which is the primary condition. The new location of a downstream process must have a sequence number greater than or equal to the workstation sequence numbers of all its preceding processes. ② Optimal load, to ensure the correction process does not destroy existing optimization results. Among all candidate workstations that meet the "timing feasibility" criterion, the algorithm selects the workstation with the lowest current total load (i.e., the least idle) as the target. The execution flow is as follows: a. Location conflict: The algorithm detects during inspection that a downstream process S has been assigned to workstation K. s One of its preceding processes, P, was assigned to workstation K. p And K s <K p This means that the subsequent processes are scheduled earlier; b. Generating a candidate list: The algorithm uses K... p Using this as a baseline, iterate through all workstations and retrieve all serial numbers. The workstation is added to the candidate list; c. Evaluate load: The algorithm calculates the total working hours currently allocated to each workstation in the "candidate list" (i.e., the current load). d. Selecting the optimal target: The algorithm selects the workstation with the lowest current total load from the "candidate list"; e. Execution migration: The algorithm moves the violating post-process S from the original workstation K. s Remove it from the list and assign it to the "optimal target" workstation selected in the previous step.

[0052] The above execution process can migrate conflicting processes to a new workstation that meets timing requirements and has a better workload, thereby ensuring that all individuals strictly adhere to production process constraints.

[0053] The following is a simple example illustrating how an individual violating constraints can be progressively corrected into a feasible solution: Prerequisites: Process dependency: Operation C (engine installation, 30 seconds) must be completed before operation D (wire harness connection, 15 seconds); Production line layout: 3 workstations in total, flowing in sequence (unit 1 -> unit 2 -> unit 3).

[0054] Offspring individuals: Unit 1 (load: 40s): [Operation P, Operation Q]; Unit 2 (load: 55s): [Operation R, Operation S]; Unit 3 (load: 25s): [Operation T].

[0055] Now, process C is assigned to unit 2, and process D is assigned to unit 1, resulting in: Unit 1 (load: 55s): [Operation P, Operation Q, Operation D]; Unit 2 (load: 85s): [Operation R, Operation S, Operation C]; Unit 3 (load: 25s): [Operation T].

[0056] The corresponding correction steps are as follows: Step 1: Detect conflict and initiate a feasibility check. When it traverses the dependency {Process C->Process D}, it finds that the predecessor process C is in unit 2 (K). p =2), the subsequent process D is in unit 1 (K)s =1). A collision was detected because K s <K p (1<2), the subsequent process is placed before the preceding process; Step 2: Locate the violating process and generate a candidate list. The algorithm locates the violating process as the subsequent process D, whose preceding process C is in unit 2. Therefore, the new position of D must be in or after "unit 2", and the temporally feasible candidate list is {unit 2, unit 3}.

[0057] Step 3: Evaluate the load of candidate stations. The algorithm checks the current load of each station in the candidate list (at this point, C is in unit 2, D is still in unit 1, and the load has not changed): Current load of Unit 2 (including process C): 85s Current load of Unit 3: 25s Comparison load: 25s (Unit 3) < 85s (Unit 2).

[0058] Step 4: Select the optimal target and execute the migration. Based on the "load optimal" criterion, the algorithm selects unit 3 as the new workstation for process D and moves process D (15 seconds) from unit 1 to unit 3.

[0059] Step 5: Output the corrected feasible solution: Unit 1 (Load: 40s): [Operation P, Operation Q] Unit 2 (Load: 85s): [Operation R, Operation S, Operation C] Unit 3 (Load: 40s): [Process T, Process D] like Figure 5 As shown, the general process allocation scheme output by the present invention can achieve excellent comprehensive load balancing effect on all target models (when the weights are the same).

[0060] Step 5: Transfer the repaired offspring population As the next generation of new population P t+1 Determine if the current generation t has reached the set maximum value Gen. If the termination condition is met, proceed to the next step; otherwise, return to step 4 and continue iterating.

[0061] Step 6: The algorithm terminates, outputting the individual with the highest fitness value throughout the evolutionary process. The process allocation scheme corresponding to this individual is the final optimization result. This scheme is a general solution for multi-product mixed-line production scenarios, capable of achieving comprehensive load balancing across all target vehicle models and assigning a unique production unit to each process.

[0062] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for optimizing multi-model mixed-line final assembly processes based on load balancing, characterized in that, Includes the following steps: Step 1: Input the basic data for mixed-line production, including the set of production models M, the set of production processes J, the set of production units K, and the standard operating time of each model i on each process j. Set the operating parameters of the genetic algorithm, including population size N and crossover rate. Variation rate The maximum number of generations (Gen) is determined, and the running parameters are initialized. Step 2: Generate an initial population according to the encoding scheme, wherein the encoding scheme represents a process allocation scheme as a string of length [length missing]. The initial population consists of individual sequences, where each position in the sequence corresponds to a process step, and its gene value represents the production unit number to which that process step is assigned. A genetic algorithm is used to randomly generate N individual sequences that satisfy the process step dependency relationships, forming the initial population. ; Step 3: For each individual, calculate the weighted mean square error of the load across all vehicle models based on the process allocation scheme it represents. Based on this, its fitness value (Fitness) is calculated using a negative exponential function; Step 4: For the current population The selection, crossover, and mutation operations are performed sequentially to generate the offspring population. ; Step 5: Check whether the offspring individuals violate the process dependency constraints. If they do, correct them through the feasibility guarantee mechanism to ensure that all individuals are feasible solutions. Step 6: Transfer the corrected offspring population As the next generation population P t+1 Determine whether the current generation t has reached the set maximum value Gen. If not, return to step 3; if it has, terminate the algorithm and output the process allocation scheme corresponding to the individual with the highest fitness value in the entire evolution process as the final optimization result.

2. The method according to claim 1, characterized in that, The encoding scheme is specifically as follows: the individual sequence is represented as... ... ,in , Index i represents the process number, gene value This indicates the production unit number to which the process is assigned.

3. The method according to claim 1, characterized in that, The following constraints must be satisfied when generating the initial population in step 2 and when correcting infeasible solutions in step 5: Process dependency constraint: For any vehicle model If the process Must be in the process If completed previously, then the following conditions must be met. ,in Let be the completion time of process j. It is the process of model i Standard operating time; Process unique assignment constraint: Each process must and can only be assigned to one production unit, that is, for any process... ,have ,in A binary variable representing a process. Whether it has been assigned to production unit k; Multi-model consistency constraint: All models use the same process-unit mapping structure, that is, for any model ,have .

4. The method according to claim 1, characterized in that, The calculation steps for the fitness value (Fitness) mentioned in step 3 are as follows: ① For the current individual sequence, iterate through each vehicle type i∈M; ② Calculate the total working hours of vehicle model i in production unit k. ; ③ Calculate the average total working hours of vehicle model i across all production units. ; ④ Calculate the load mean square error of vehicle model i ; ⑤ Calculate the weighted global comprehensive evaluation index ,in Let be the weighting coefficient for vehicle model i, and ; ⑥ Calculate the fitness function where α is a constant; Wherein, the weighting coefficient It can be set according to changes in vehicle model business needs and market demand.

5. The method according to claim 1, characterized in that, The selection operation described in step 4 employs a strategy combining elite retention and roulette wheel betting, specifically including: Elite selection: The top 10% of individuals in the current population based on their fitness values ​​are directly copied to the next generation of the population; Roulette wheel selection: From the remaining population after removing elite individuals, the selection probability is calculated based on the proportion of individual fitness value to total fitness, and individuals that account for 90% of the population size are selected by roulette wheel selection.

6. The method according to claim 1, characterized in that, The specific process of the crossover operation described in step 4 is as follows: For each production unit k, calculate its comprehensive load standard deviation across all vehicle models in both parent individuals. ; Compare the combined load standard deviation of the same production unit k in the two parent generations. Inheritance in offspring has a smaller chance of success. The process combination scheme of this unit in the parent individual of the value; The formula for calculating the comprehensive load standard deviation is as follows: , It is the total man-hours of model i in production unit k. For unit Weighted average working hours when carrying all different vehicle models.

7. The method according to claim 1, characterized in that, The specific process of the mutation operation described in step 4 is to perform an exchange mutation operation on the offspring individuals generated after crossover at a preset mutation rate, and randomly exchange the production units allocated by the two processes.

8. The method according to claim 1, characterized in that, After the crossover and mutation operations in step 4, the process integrity conflict detection and repair is also included: detecting whether there are duplicate or missing processes in the offspring individuals, and ensuring that each process is uniquely assigned by reallocation.

9. The method according to claim 1, characterized in that, The feasibility guarantee mechanism described in step 5 addresses cross-workstation conflicts by using a global relocation strategy to migrate the conflicting process to a new workstation that meets timing constraints and has a better load.

10. The method according to claim 9, characterized in that, The global relocation strategy includes the following steps: a. Locate the conflicting processes and the workstations containing their preceding processes; b. Generate a list of all candidate workstations that meet the timing constraints; c. Assess the current load of the candidate workstations; d. Select the candidate workstation with the lowest load as the target; e. Move the conflicting process to the target workstation.

Images