Flexible job shop joint scheduling optimization method for multiple types of AGVs (Automatic Guided Vehicles)

Abstract

The present invention relates to an optimization method for integrated joint scheduling of production and logistics in a flexible job shop (FJSP) having a plurality of different types of automated guided vehicles (AGVs). The invention belongs to the field of assembly workshop production scheduling. Comprising the following steps: 1) according to a special assembly workshop machine and AGV combined scheduling process, Tent chaotic mapping is adopted to initialize a scheduling scheme and encode the scheduling scheme; 2) performing iterative optimization adjustment on the scheduling scheme through an improved multi-target artificial bee colony algorithm; and 3) carrying out production scheduling by using the optimized scheduling scheme. According to the method, the maximum completion time and the total energy consumption are optimized at the same time, the production efficiency is concerned, the requirements of green manufacturing and sustainable development are also considered, and enterprises are helped to achieve cost reduction and efficiency improvement, especially in the production process sensitive to energy consumption.

Description

Technical Field

[0001] This invention relates to an optimized method for integrated scheduling of production and logistics in a flexible job shop (FJSP) with various types of automated guided vehicles (AGVs). This invention belongs to the field of assembly shop production scheduling. Background Technology

[0002] As modern industry shifts towards flexible production models characterized by small batches and diverse product varieties, the complexity and importance of production scheduling are becoming increasingly prominent. This is particularly true in high-end manufacturing sectors such as aerospace, defense, and fine chemicals, where specialized assembly workshops play a crucial role. These workshops typically handle high-value, complex materials with unique physical or chemical properties (e.g., high energy, volatility, environmental sensitivity), placing extremely stringent requirements on safety, stability, cycle time coordination, and timeliness. Any delay or error in material transport can lead to serious quality problems or even safety accidents. To improve production accuracy and safety in these specialized environments, automated guided vehicles (AGVs) have become key equipment for achieving precise, efficient, and automated material handling. However, traditional production scheduling and AGV logistics scheduling are often separate, and existing research largely ignores AGV transport time or simplifies it as a fixed portion of processing time. This simplification severely impacts the practical feasibility of scheduling schemes because processing and transport are closely coupled: the completion time of preceding processing tasks determines the start of transport tasks, and transport duration directly affects the start time of subsequent processes.

[0003] Furthermore, existing studies on flexible job shop scheduling (FJSP-T) with AGV transportation largely assume that all AGVs are homogeneous, neglecting the combined impact of AGV heterogeneity (such as different speeds, energy consumption, load capacities, and even explosion-proof ratings) on total energy consumption (TEC) and maximum time to complete (Makespan). In real-world manufacturing environments, workshops often deploy heterogeneous fleets of various AGV models to meet diverse task requirements. Ignoring these differences leads to inaccurate estimates of total energy consumption and misses opportunities to find a better balance between production efficiency and energy sustainability.

[0004] To address this problem, current mainstream solutions are divided into exact algorithms and approximate algorithms. While exact algorithms (such as Mixed Integer Linear Programming (MILP) and Constrained Programming (CP) can guarantee optimal solutions, they are only suitable for small-scale problems and are computationally infeasible for large-scale instances. Therefore, heuristic and metaheuristic algorithms (such as Artificial Bee Colony Algorithm ABC) are widely used to solve complex scheduling problems due to their ability to find high-quality feasible solutions within acceptable timeframes. In recent years, "learning-based" methods that combine reinforcement learning (RL) with heuristic algorithms have significantly improved solution performance and generalization ability by endowing algorithms with the ability to learn from experience. Therefore, developing a joint scheduling method that fully considers the heterogeneity of AGVs and is suitable for the high-requirement environment of special assembly workshops has significant theoretical and practical implications. Summary of the Invention

[0005] This invention aims to address the shortcomings of existing technologies in considering the heterogeneity of AGVs and their inability to meet the complex scheduling requirements of specialized assembly workshops. It provides a flexible workshop joint scheduling method for multiple types of AGVs that can simultaneously optimize maximum completion time and total energy consumption. This method generates scheduling schemes that are more practical and balance production efficiency with energy conservation.

[0006] The technical solution adopted by the present invention to achieve the above objectives is as follows:

[0007] A flexible job shop joint scheduling optimization method for multiple types of AGVs includes the following steps:

[0008] 1) Based on the joint scheduling process of machines and AGVs in the special assembly workshop, the scheduling scheme is initialized using Tent chaotic mapping and the scheduling scheme is encoded;

[0009] 2) The scheduling scheme is iteratively optimized and adjusted by improving the multi-objective artificial bee colony algorithm;

[0010] 3) Use the optimized scheduling scheme for production scheduling.

[0011] Step 1) includes the following steps:

[0012] 1.1) Initialize the workshop environment information and the transfer time matrix between machines. The workshop environment information includes the workpiece set, the process sequence of each workpiece, the machine set, and information on the multi-type AGV fleet.

[0013] 1.2) A scheduling scheme consisting of process sequencing, machine allocation, and AGV scheduling is represented using a process-based three-layer chromosome encoding method;

[0014] 1.3) The initialization information is processed using the Tent chaotic mapping to generate P individuals consisting of three layers of chromosomes, i.e., P scheduling schemes;

[0015] 1.4) By minimizing the dominance relations, the initial population P0 = {x1, x2, x3, ... x} is initialized. P This is classified as a Pareto front.

[0016] Step 1.2) specifically refers to:

[0017] Process ordering chromosome OS: An integer vector where gene values ​​represent workpiece numbers, and the frequency of identical integers indicates the process number for that workpiece. The order of genes in the vector defines the global processing order for all processes.

[0018] Machine allocation chromosome MA: An integer vector with the same length as the OS chromosome. The gene value at the k-th position on the MA chromosome indicates that the k-th process on the OS chromosome is assigned to the m-th machine in its candidate machine set for processing.

[0019] AGV scheduling chromosome AD: An integer vector whose length is equal to the total number of transport tasks. The gene value in the vector represents the unique number of the AGV assigned to perform the corresponding transport task. This number is used to characterize the type of AGV and its index number in that type.

[0020] Step 1.3) specifically refers to:

[0021] Choose an initial value x0∈(0,1) and set a warm-up period B. Use the chaotic value iteration formula to iteratively generate several chaotic values ​​step by step, skipping the initial "warm-up period B" values, and starting from B+1, use the final chaotic value sequence {u1,u2…u...u...}. n The chaotic value sequence is mapped to each decision variable represented by a chromosome in the population to generate an initial population P0 of size P. The chaotic value iteration formula is:

[0022]

[0023] in, This represents the value of the chaotic variable generated in the (t+1)th iteration. This represents the value of the chaotic variable generated in the t-th iteration, and 'a' represents a control parameter used to control the specific distribution of the chaotic sequence.

[0024] Step 1.4) specifically refers to:

[0025] For any individual p,q∈P, the relation in which p dominates q is:

[0026]

[0027] Among them, C max(·) indicates the maximum completion time, and TEC(·) indicates the total energy consumption;

[0028] For each individual p∈P, define the dominance set S(p) and the dominance number n(p):

[0029] S(p)={q∈P|p<q},n(p)=|{q∈P|p<q}|

[0030] Let the first frontier be F1 = {p ∈ P | n(p) = 0}; subsequent frontiers are "stripped" of F1 through iteration. k Obtain, that is, for all p∈F k Given q∈S(p), assign n(q)-1 to n(q) to generate a partition with a non-dominant level.

[0031] Step 2) includes the following steps:

[0032] 2.1) Set the control parameters of the multi-objective artificial bee colony algorithm and obtain the optimal parameter combination through Taguchi experiments. Define the fitness function of the algorithm: F(x)=(Cmax(x),TEC(x)), and define the Q-learning tuple (S,A,R,Q), where s∈S is the state, a∈A is the operator, R(s,a) is the reward, and Q(s,a) is the Q value. The reward rule is as follows: if the new solution dominates the old solution, then R=10 and it is replaced; if the new solution is dominated by the old solution, then R=0; if they do not dominate each other, then R=5 and the new solution is added to the archive of non-dominated archive Ω.

[0033] 2.2) For each individual in the initial population, decode its three chromosome layers, calculate its corresponding maximum completion time Cmax(x) and total energy consumption TEC(x), and perform fast non-dominated sorting on the entire population to stratify all individuals into different Pareto fronts F1, F2, ..., F k ;

[0034] 2.3) Treating individuals as mercenary bees, for each mercenary bee, i.e., individual x i Perform the bee-hiring phase;

[0035] 2.4) Treat each individual as an observation bee and perform the observation bee phase for each observation bee;

[0036] 2.5) Treat each individual as a scout bee and perform the scout bee phase for each observation bee;

[0037] 2.6) Increment the current iteration count by 1, and determine if the current iteration count has reached the maximum number of iterations, MaxIter. If it has, stop the iteration and output the current scheduling scheme; otherwise, return to step 2.2).

[0038] Step 2.3) includes the following steps:

[0039] 2.3.1) Action selection: Based on the current solution state s t An ε-greedy strategy is used to select a search operator a from the action space A. t That is, the action with the maximum Q value is selected with a probability of 1-ε, and an action is randomly selected with an exploration rate of ε.

[0040] 2.3.2) Generating candidate solutions: using the selected operator a t For individual x i Perform a neighborhood search to generate a candidate solution x′;

[0041] 2.3.3) Evaluation and Reward: Calculate the fitness F(x′) of x′ and compare it with F(x′). i Perform a Pareto comparison to determine the reward value R. t ;

[0042] 2.3.4) Greedy selection and update: If F(x′) is not inferior to F(x′), then... i If the value is either dominant or non-dominated, then replace x with x′. i ;

[0043] 2.3.5) Q-value update: based on the reward R obtained. t and the new state s after the transition t+1 Update the Q-table using the Bellman equation:

[0044] Q(S t ,a t )=Q(S t ,a t )+α[r t+1 +γmaxQ(S t+1 ,:)-Q(S t ,a t )]

[0045] Where α is the learning rate and γ is the discount factor.

[0046] Step 2.4) includes the following steps:

[0047] 2.4.1) Individual selection: The fitness of each individual is used to calculate its search probability, and the roulette wheel selection method is used to select individuals for neighborhood search;

[0048] 2.4.2) Applying variable neighborhood search: N1-N55 neighborhood structures are sequentially used to search the selected individual x. k The perturbation is performed until a better solution is found. The neighborhood structure is specifically as follows:

[0049] N1: On the OS chromosome, randomly select position i and swap it with the operation at position i+2;

[0050] N2: On the OS chromosome, randomly select position i and insert the process on it before position i-3;

[0051] N3: On the OS chromosome, randomly select position i and insert the process on it after position i+3;

[0052] N4: Randomly select a process and reselect a different, feasible processing machine for it on the MA chromosome;

[0053] N5: Randomly select a transportation task and reassign it a different, feasible AGV on the AD chromosome;

[0054] 2.4.3) Acceptance and Update: If the fitness is improved, the search starts again from the first neighborhood; otherwise, the search continues to the next neighborhood until the number of searches reaches the number of neighborhood searches for the observation bee, LS.

[0055] Step 2.5) specifically refers to:

[0056] If an individual x1 fails to improve its fitness in N consecutive iterations during the observation bee phase, it is determined that the individual has stagnated. In this case, the scout bee mechanism is activated, and individual x1 is abandoned. l And a completely new random individual is generated to replace it using a chaotic initialization method.

[0057] Step 3) includes the following steps:

[0058] 3.1) Merge all non-dominated solutions generated in the current iteration with the external Pareto archive A*, sort the merged set by non-dominated order, and use the new set of non-dominated solutions as the updated archive Ω.

[0059] 3.2) Output the Pareto archive Ω of the optimal scheduling scheme that is non-dominated between the two objectives of maximum completion time and total energy consumption, and use the scheduling scheme for production scheduling.

[0060] The present invention has the following beneficial effects and advantages:

[0061] 1. Closer to reality: This invention is the first to incorporate the type differences (heterogeneity) of AGVs into the joint scheduling model of FJSP, making the scheduling scheme more consistent with the complex situation of real workshops, especially suitable for special assembly environments with special requirements for equipment, thus improving the feasibility and practicality of the scheme.

[0062] 2. Multi-objective optimization: By simultaneously optimizing maximum completion time and total energy consumption, this invention not only focuses on production efficiency but also takes into account the requirements of green manufacturing and sustainable development, helping enterprises to reduce costs and increase efficiency, especially in energy-sensitive production processes.

[0063] 3. Superior Algorithm Performance: The proposed Q-learning hybrid multi-objective artificial bee colony algorithm forms a collaborative optimization mechanism of "global intelligent exploration - local in-depth mining - diversity maintenance" through Q-learning-based adaptive exploration in the hired bee phase, VNS-based elite solution depth optimization in the observer bee phase, and a restart strategy for scout bees. This mechanism has a clear division of labor and advanced strategies, exhibiting stronger optimization capabilities and higher solution efficiency when solving complex FJSP-MTA problems. Attached Figure Description

[0064] Figure 1 This is a general flowchart of the embodiments of the present invention;

[0065] Figure 2 This is a schematic diagram of the three-layer coding scheme of the present invention;

[0066] Figure 3 This is an optimized Gantt chart from one of the examples in this invention;

[0067] Figure 4 This is a multi-objective Pareto diagram of the results of this invention. Detailed Implementation

[0068] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0069] like Figure 1 As shown, a flexible workshop joint scheduling optimization method for multiple types of AGVs is described. Specifically, it includes the following steps:

[0070] Step 1: Initialize 60 scheduling schemes and encode them;

[0071] Step 2: Initialize algorithm parameters and iteratively optimize and adjust the scheduling scheme by improving the multi-objective artificial bee colony algorithm;

[0072] Step 3: Select the optimized joint scheduling scheme for assembly workshop machines and AGVs for production scheduling;

[0073] Step 1.1: Initialize Problem Parameters: This embodiment considers a special assembly workshop containing 4 machines (M1, M2, M3, M4) and 3 AGVs of 2 types (AGV_T1, AGV_T2a, AGV_T2b), which needs to complete a production order containing 4 workpieces (J1, J2, J3, J4), with each workpiece containing 3 processes. In the workshop layout, M0 is the material loading / unloading and finished product area.

[0074] Table 1 Machine Energy Consumption Table

[0075] parameter Machine 1 (M1) Machine 2 (M2) Machine 3 (M3) <![CDATA[Processing power (PP k )]]> 2.5 2.1 2.7 <![CDATA[Idle Power (IP k )]]> 1.5 1.2 1.6

[0076] Table 2 AGV Energy Consumption Table

[0077] parameter AGV Type 1 (T1) AGV Type 2 (T2) <![CDATA[Load power (APt load )]]> 2.4 1.2 <![CDATA[No-load power (APt empty )]]> 1.6 0.8

[0078] Table 3 Workpiece Processing Schedule (Unit: Time Unit, "—" indicates that the process cannot be processed on this machine)

[0079]

[0080]

[0081] Table 4 AGV Transportation Schedule (Unit: Time Unit)

[0082] AGV_T1 M0 M1 M2 M3 AGV_T2 M0 M1 M2 M3 M0 0 20 30 40 M0 0 40 60 80 M1 20 0 20 30 M1 40 0 40 60 M2 30 20 0 20 M2 60 40 0 40 M3 40 30 20 0 M3 80 60 40 0

[0083] Step 1.2: A complete scheduling scheme is represented using a three-layer chromosome coding method based on work processes. This coding consists of three parts: work process ordering (OS), machine allocation (MA), and AGV scheduling (AD), as follows: Figure 2 As shown:

[0084] a) Operation Sequencing (OS): Each workpiece has 2 operations, and a total of 6 operations form a 6-dimensional vector.

[0085] b) Machine-assigned chromosome (MA): An integer vector with the same length as the OS chromosome, which is a 6-dimensional vector.

[0086] c) AGV Scheduling Chromosome (AD): An integer vector whose length equals the total number of transport tasks. Each workpiece requires AGV transport for two processes: from the material loading / unloading area to the first machine and from the last process to the finished product area. There are a total of 12 transport tasks, and the AD vector is 12-dimensional.

[0087] Combine the three layers of chromosomes into a 24-dimensional vector in sequence.

[0088] Step 1.3: Based on the initialization information in Step 1.1, generate 60 individuals, i.e., 60 scheduling schemes, using the Tent chaotic mapping. Specifically, select an initial value x0∈(0,1) and set a warm-up period B, then iteratively generate the scheme using the formula below. Skip the initial "warm-up period B" values, and start from B+1 as the final chaotic value sequence {u1, u2…u ... n Then, by mapping this sequence to each decision variable in the population, such as the operation sequence OS, machine allocation MA, and automated guided vehicle scheduling AD, an initial population of size 60 can be generated. The generation formula is as follows:

[0089]

[0090] The initial 60 individuals are shown in Table 5:

[0091] Table 5 Initialized 60 individuals

[0092]

[0093] Step 1.4: Minimize the dominance relations to initialize the population P0 = {x1, x2, x3, ... x}. 60 The area is divided into Pareto fronts. For any p, q ∈ P, the relation that p governs q is as follows:

[0094]

[0095] For each p∈P, the dominance set and dominance number are defined in the following equation:

[0096] S(p)={q∈P|p<q},n(p)=|{q∈P|p<q}|

[0097] Where S(p) is the dominance set, representing the set of all individuals q dominated by individual p in the entire population P, and n(p) is the dominance number, representing the total number of individuals dominated by individual p. The first front is F1 = {p∈P|n(p) = 0}; subsequent fronts are formed by iteratively "stripping" F1 from the dominance set. k Obtain: For all p∈F k Given q∈S(p), n(q)←n(q)-1, a partition with a non-dominant rank is generated, in practice, the frontier rank Fk guides survivor selection.

[0098] Step 2 includes the following steps:

[0099] Step 2.1: Set the control parameters of the Q-HMOABC algorithm, including population size ps, maximum number of iterations MaxIter, number of hired bees, observer bees, and scout bees, threshold N for scout bees to trigger reset in the unimproved generation, number of neighborhood searches by observer bees LS, and learning rate α, discount factor γ, and exploration rate ε for the hired bee stage. The optimal parameter combination is obtained through Taguchi experiments, including the following six parameters: ps, N, LS, α, γ, and ε. Each parameter has three levels, with the following values: ps = {40, 60, 80}, α = {0.05, 0.1, 0.15}, γ = {0.7, 0.8, 0.9}, ε = {0.3, 0.5, 0.7}, N = {20, 30, 40}, and LS = {5, 10, 15}.

[0100] In this parameter calibration experiment, we used an L16(4) 4An orthogonal array was used. To ensure fairness, each parameter configuration was run independently 20 times. The maximum number of iterations was set to G = 150. Through comprehensive observation (analysis of the main effect plot and signal-to-noise ratio plot), the optimal parameter values ​​were finally determined as follows: ps = 60, α = 0.10, γ = 0.7, ε = 0.7, N = 20, and LS = 15.

[0101] Define the fitness function: F(x) = (Cmax(x), TEC(x)). Define the Q-learning tuple (S, A, R, Q), where s ∈ S is the state, a ∈ A is the operator, R(s, a) is the reward, and Q(s, a) is the Q-value. The reward rules are as follows: if the new solution dominates the old solution, then R = 10 and the solution is replaced; if the new solution is dominated by the old solution, then R = 0; if they do not dominate each other, then R = 5 and the new solution is added to the archive of non-dominated archives Ω.

[0102] Define the fitness function: F(x) = (Cmax(x), TEC(x)). Define the Q-learning tuple (S, A, R, Q), where s ∈ S is the state, a ∈ A is the operator, R(s, a) is the reward, and Q(s, a) is the Q-value. The reward rules are as follows: if the new solution dominates the old solution, then R = 10 and the solution is replaced; if the new solution is dominated by the old solution, then R = 0; if they do not dominate each other, then R = 5 and the new solution is added to the archive of non-dominated archives Ω.

[0103] Step 2.2: If the current iteration count has reached the maximum number MaxIter, stop the iteration, output the scheduling scheme, and continue to step 3; otherwise, continue to step 2.3.

[0104] Step 2.3: For each individual in the initial population, decode its three chromosome layers and calculate its corresponding two objective function values: maximum completion time Cmax(x) and total energy consumption TEC(x). Subsequently, perform fast non-dominated sorting on the entire population, stratifying all individuals to different Pareto fronts F1, F2, ..., F.

[0105] Step 2.4: Perform the mercenary bee phase, for each mercenary bee (individual x) i Perform the following operations in sequence:

[0106] 1) Action selection: Based on the current solution state s t An ε-greedy strategy is used to select a search operator a from the action space A. t That is, select the action with the largest Q value with a probability of 1-ε, and randomly select an action with a probability of ε.

[0107] 2) Generate candidate solutions: Apply the selected operator a t For individual x i Perform a neighborhood search to generate a candidate solution x'.

[0108] 3) Evaluation and Reward: Calculate the fitness F(x′) of x′ and compare it with F(x′). i Perform a Pareto comparison and determine the reward value R according to the rules defined in section 1.4. t .

[0109] 4) Greedy selection and update: If F(x′) is not inferior to F(x′), then... i (i.e., dominance or non-dominance), then replace x with x′. i .

[0110] 5) Q-value update: based on the reward R obtained and the new state s after the transition. +1 Update the Q table using the following Bellman equation:

[0111] Q(S t ,a t )=Q(S t ,a t )+α[r t+1 +γmaxQ(S t+1 ,:)-Q(S t ,a t )]

[0112] Step 2.5: Execute the observation bee phase of the algorithm. Specifically, perform the following operations sequentially for each observation bee:

[0113] 1) Individual selection: The fitness of each individual is calculated as its search probability, and then a roulette wheel selection method is used to select individuals for neighborhood search. The higher the fitness of an individual, the greater its probability of being selected.

[0114] 2) Apply Variable Neighborhood Search (VNS): sequentially search the selected individuals x k The following five carefully designed neighborhood structures are applied to perturb the solution until a better solution is found:

[0115] N1(swap2): On the OS chromosome, randomly select position i and swap it with the operation at position i+2.

[0116] N2(insert3B): On the OS chromosome, randomly select position i and insert the process therein before position i-3.

[0117] N3(insert3F): On the OS chromosome, randomly select position i and insert the process thereon after position i+3.

[0118] N4 (changeM): Randomly select a process and reselect a different, feasible processing machine on the MA chromosome.

[0119] N5 (changeA): Randomly select a transportation task and reassign it to a different, feasible AGV on the AD chromosome.

[0120] 3) Acceptance and Update: Once an improvement is made, the search will restart from the first neighborhood; otherwise, it will continue to the next neighborhood until the number of searches reaches LS.

[0121] Step 2.6: The scout bee phase of the algorithm is specifically performed as follows: Each individual in the population is checked. If an individual x... l If an individual's fitness does not improve in N consecutive iterations, it is determined that the individual has stagnated. At this point, the scout bee mechanism is activated, and individual x is abandoned. l And a completely new random individual is generated to replace it using the chaotic initialization method in step 1.3, thereby maintaining the diversity of the population.

[0122] Step 2.7: Increment the current iteration count by 1, and return to step 2.2.

[0123] Step 3 is described in detail below:

[0124] Step 3.1: Merge all non-dominated solutions generated in the current iteration with the external Pareto archive, then sort the merged set by non-dominated order, and use the new set of non-dominated solutions as the updated archive Ω.

[0125] Step 3.2: Based on actual production needs, select an optimized joint scheduling scheme for assembly workshop machines and AGVs for production scheduling.

[0126] The experimental results obtained by using this method for workshop scheduling are as follows: Figure 3 and Figure 4 As shown.

Claims

1. A flexible workshop joint scheduling optimization method for multiple types of AGVs, characterized in that, Includes the following steps: 1) Based on the joint scheduling process of machines and AGVs in the special assembly workshop, the scheduling scheme is initialized using Tent chaotic mapping and the scheduling scheme is encoded; 2) The scheduling scheme is iteratively optimized and adjusted by improving the multi-objective artificial bee colony algorithm; 3) Use the optimized scheduling scheme for production scheduling.

2. The method for joint scheduling optimization of flexible workshops for multiple types of AGVs according to claim 1, characterized in that, Step 1) includes the following steps: 1.1) Initialize the workshop environment information and the transfer time matrix between machines. The workshop environment information includes the workpiece set, the process sequence of each workpiece, the machine set, and information on the multi-type AGV fleet. 1.2) A scheduling scheme consisting of process sequencing, machine allocation, and AGV scheduling is represented using a process-based three-layer chromosome encoding method; 1.3) The initialization information is processed using the Tent chaotic mapping to generate P individuals consisting of three layers of chromosomes, i.e., P scheduling schemes; 1.4) By minimizing the dominance relations, the initial population P0 = {x1, x2, x3, ... x} is initialized. P This is classified as a Pareto front.

3. The method for joint scheduling optimization of flexible workshops for multiple types of AGVs according to claim 2, characterized in that, Step 1.2) specifically refers to: Process ordering chromosome OS: An integer vector where gene values ​​represent workpiece numbers, and the frequency of identical integers indicates the process number for that workpiece. The order of genes in the vector defines the global processing order for all processes. Machine allocation chromosome MA: An integer vector with the same length as the OS chromosome. The gene value at the k-th position on the MA chromosome indicates that the k-th process on the OS chromosome is assigned to the m-th machine in its candidate machine set for processing. AGV scheduling chromosome AD: An integer vector whose length is equal to the total number of transport tasks. The gene value in the vector represents the unique number of the AGV assigned to perform the corresponding transport task. This number is used to characterize the type of AGV and its index number in that type.

4. The method for joint scheduling optimization of flexible workshops for multiple types of AGVs according to claim 2, characterized in that, Step 1.3) specifically refers to: Choose an initial value x0∈(0,1) and set a warm-up period B. Use the chaotic value iteration formula to iteratively generate several chaotic values ​​step by step, skipping the initial "warm-up period B" values, and starting from B+1, use the final chaotic value sequence {u1, u2, ... u...}. n The chaotic value sequence is mapped to each decision variable represented by a chromosome in the population to generate an initial population P0 of size P. The chaotic value iteration formula is: in, This represents the value of the chaotic variable generated in the (t+1)th iteration. represents the chaotic variable value generated in the t-th iteration, and 'a' represents a control parameter used to control the specific distribution of the chaotic sequence.

5. The method for joint scheduling optimization of flexible workshops for multiple types of AGVs according to claim 2, characterized in that, Step 1.4) specifically refers to: For any individual p,q∈P, the relation in which p dominates q is: Among them, C max (·) indicates the maximum completion time, and TEC(·) indicates the total energy consumption; For each individual p∈P, define the dominance set S(p) and the dominance number n(p): Let the first frontier be P1 = {p ∈ P | n(p) = 0}; subsequent frontiers are obtained by iteratively "stripping" F. k Obtain, that is, for all p∈F k Given q∈S(p), assign n(q)-1 to n(q) to generate a partition with a non-dominant level.

6. The method for joint scheduling optimization of flexible workshops for multiple types of AGVs according to claim 1, characterized in that, Step 2) includes the following steps: 2.1) Set the control parameters of the multi-objective artificial bee colony algorithm and obtain the optimal parameter combination through Taguchi experiments. Define the fitness function of the algorithm: F(x)=(Cmax(x),TEC(x)), and define the Q-learning tuple (S,A,R,Q), where s∈S is the state, a∈A is the operator, R(s,a) is the reward, and Q(s,a) is the Q value. The reward rule is as follows: if the new solution dominates the old solution, then R=10 and it is replaced; if the new solution is dominated by the old solution, then R=0; if they do not dominate each other, then R=5 and the new solution is added to the archive of non-dominated archive Ω. 2.2) For each individual in the initial population, decode its three chromosome layers, calculate its corresponding maximum completion time Cmax(x) and total energy consumption TEC(x), perform fast non-dominated sorting on the entire population, and stratify all individuals to different Pareto fronts F1, F2, ..., F k ; 2.3) Treating individuals as mercenary bees, for each mercenary bee, i.e., individual x i Perform the bee-hiring phase; 2.4) Treat each individual as an observation bee and perform the observation bee phase for each observation bee; 2.5) Treat each individual as a scout bee and perform the scout bee phase for each observation bee; 2.6) Increment the current iteration count by 1, and determine if the current iteration count has reached the maximum number of iterations, MaxIter. If it has, stop the iteration and output the current scheduling scheme; otherwise, return to step 2.2).

7. The method for joint scheduling optimization of flexible workshops for multiple types of AGVs according to claim 6, characterized in that, Step 2.3) Includes the following steps: 2.3.1) Action selection: Based on the current solution state s t An ε-greedy strategy is used to select a search operator a from the action space A. t That is, the action with the maximum Q value is selected with a probability of 1-ε, and an action is randomly selected with an exploration rate of ε. 2.3.2) Generating candidate solutions: using the selected operator a t For individual x i Perform a neighborhood search to generate a candidate solution x′; 2.3.3) Evaluation and Reward: Calculate the fitness F(x′) of x′ and compare it with F(x′). i Perform a Pareto comparison to determine the reward value R. t ; 2.3.4) Greedy selection and update: If F(x′) is not inferior to F(x′), then... i If the value is either dominant or non-dominated, then replace x with x′. i ; 2.3.5) Q-value update: based on the reward R obtained. t and the new state s after the transition t+1 Update the Q-table using the Bellman equation: Q(S t ,a t )=Q(S t ,a t )+α[r t+1 +γmaxQ(S t+1 ,:)-Q(S t ,a t )] Where α is the learning rate and γ is the discount factor.

8. The method for joint scheduling optimization of flexible workshops for multiple types of AGVs according to claim 6, characterized in that, Step 2.4) Includes the following steps: 2.4.1) Individual selection: The fitness of each individual is used to calculate its search probability, and the roulette wheel selection method is used to select individuals for neighborhood search; 2.4.2) Applying variable neighborhood search: N1-N55 neighborhood structures are sequentially used to search the selected individual x. k The perturbation is performed until a better solution is found. The neighborhood structure is specifically as follows: N1: On the OS chromosome, randomly select position i and swap it with the operation at position i+2; N2: On the OS chromosome, randomly select position i and insert the process on it before position i-3; N3: On the OS chromosome, randomly select position i and insert the process on it after position i+3; N4: Randomly select a process and reselect a different, feasible processing machine for it on the MA chromosome; N5: Randomly select a transportation task and reassign it a different, feasible AGV on the AD chromosome; 2.4.3) Acceptance and Update: If the fitness is improved, the search starts again from the first neighborhood; otherwise, the search continues to the next neighborhood until the number of searches reaches the number of neighborhood searches for the observation bee, LS.

9. A joint scheduling optimization method for flexible workshops with multiple types of AGVs according to claim 6, characterized in that, Step 2.5) specifically refers to: If a certain individual x l If an individual's fitness does not improve in N consecutive iterations during the observation bee phase, it is determined that the individual has stagnated. At this point, the scout bee mechanism is activated, and individual x is abandoned. l And a completely new random individual is generated to replace it using a chaotic initialization method.

10. The method for joint scheduling optimization of flexible workshops for multiple types of AGVs according to claim 1, characterized in that, Step 3) includes the following steps: 3.1) Merge all non-dominated solutions generated in the current iteration with the external Pareto archive A*, sort the merged set by non-dominated order, and use the new set of non-dominated solutions as the updated archive Ω. 3.2) Output the Pareto archive Ω of the optimal scheduling scheme that is non-dominated between the two objectives of maximum completion time and total energy consumption, and use the scheduling scheme for production scheduling.

Images