A method and apparatus for scheduling a sirru system in a worker transfer scenario
The Sairu system scheduling method optimized by dual-population collaborative search and Q-learning solves the problem of efficient scheduling in worker transfer scenarios, achieving efficient and intelligent scheduling and improving production efficiency and flexible response capabilities.
Patent Information
- Application Number
- CN202310674957.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-08
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2043-06-08
AI Technical Summary
Existing SERROR system scheduling algorithms involve large computational loads in worker transfer scenarios, making them difficult to solve efficiently. Furthermore, existing intelligent optimization algorithms have slow convergence speeds and cannot guarantee the quality of solutions.
A dual-population collaborative approach based on probabilistic learning and iterative greedy search is adopted, which combines the selection mechanism of Q learning and local search to optimize worker-batch, worker-task, batch-slave allocation and batch sorting. Efficient scheduling is achieved through population initialization, dual-population collaborative search, local search and probabilistic model update.
It achieves efficient solution to the scheduling problem of the SERUS system in the scenario of worker transfer, improves the solution efficiency and solution quality, and can better leverage the flexibility of the SERUS system to improve production efficiency and quickly respond to personalized consumer needs.
Smart Images

Figure CN116796963B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of semaphore system scheduling technology, and in particular to a semaphore system scheduling method and apparatus based on worker transfer scenarios. Background Technology
[0002] With economic globalization and the rapid development of emerging technologies, the business environment faced by the manufacturing industry has undergone significant changes. Consumer demand is becoming increasingly diversified and personalized, product lifecycles are short, and technology is rapidly evolving, exhibiting characteristics of high variety and small batch production. How to quickly respond to customer needs under the C2B business model is crucial for enterprises to enhance their competitiveness. Against this backdrop, the Seru production system (SPS), originating from Japanese electronics assembly companies, can improve the flexibility of the production system through flexible worker reorganization, quickly meeting personalized consumer demands. SPS comprises multiple production units (called Seru), each containing one or more workers, with easily movable equipment and workbenches. Therefore, assembly units in SPS can be built, disassembled, and adjusted according to changes in customer demand. During production, the flexible adjustment of workers within a unit can optimize product processing capabilities and improve the system's responsiveness to changes in demand. Research shows that SPS not only brings economic benefits to enterprises but also increases worker autonomy and motivation, reduces the number of workers, reduces inventory, and lowers material consumption and energy emissions. Due to its advantages, SPS has been successfully applied by many companies, such as Sony, Canon, Panasonic, NEC, and Foxconn. However, the high flexibility of SPS requires the support of scheduling optimization algorithms to ensure its feasibility and efficiency. Therefore, designing efficient scheduling methods for SPS can better leverage its flexibility, improve production efficiency, quickly meet personalized consumer needs, and help enterprises enhance their core competitiveness.
[0003] The seru system scheduling problem with worker transfer (SSSP-WT) involves four coupled subproblems: worker-batch allocation, worker-task allocation, batch-seru allocation, and batch sorting. As a complex combinatorial optimization problem, its production scheduling process is characterized by high coupling, nonlinearity, multiple minima, strong constraints, NP-hardness, and multiple objectives, posing a significant challenge to optimization. Existing scheduling optimization algorithms mainly fall into three categories: exact algorithms, heuristic algorithms, and intelligent optimization algorithms. Exact algorithms, including branch and bound, branch and price, and linear programming, can obtain the global optimum, but as the problem size increases, the computational cost becomes enormous, failing to achieve efficient problem solving. Heuristic algorithms use predefined heuristic rules to search the solution space, ensuring efficiency but not the quality of the solution. Intelligent optimization algorithms can obtain a better solution through iterative search, but traditional intelligent optimization methods mostly follow basic steps such as population initialization, crossover and mutation, and individual selection. These methods suffer from insufficient integration of problem characteristics and slow convergence speed. Summary of the Invention
[0004] The present invention aims to at least partially solve one of the technical problems in the related art.
[0005] To address this, this invention proposes a scheduling method for the Sairu system based on worker transfer scenarios. The algorithm consumes little computation, can efficiently solve the problem while ensuring both efficiency and quality, and achieves efficient and intelligent scheduling.
[0006] Another objective of this invention is to propose a scheduling device for a worker transfer system.
[0007] To achieve the above objectives, this invention proposes a scheduling method for a semaphore system based on a worker transfer scenario, comprising:
[0008] The coupled subproblems of the Selyu system scheduling problem are obtained, and a first population and a second population are constructed based on a first plurality of individuals to optimize the coupled subproblems, according to a preset encoding and decoding rule.
[0009] A probability-based search is performed on the first population and an iterative greedy search is performed on the second population. The population to be entered is selected based on the first search result and the Q-learning selection mechanism.
[0010] A second search result is obtained by performing a Q-learning-based local search on the second or more individuals in the selected population;
[0011] Based on the second search result and using the second plurality of individuals to update the probability model;
[0012] Determine whether the updated probability model meets the termination criterion. If it does, output the result.
[0013] In addition, the Sairu system scheduling method based on worker transfer scenarios according to the above embodiments of the present invention may also have the following additional technical features:
[0014] Further, in one embodiment of the present invention, the termination criterion is CPU runtime; the coupled subproblems include worker-batch allocation, worker-task allocation, batch-Syrian allocation, and batch sorting; the encoding / decoding rules include a four-column encoding structure and decoding rules; wherein,
[0015] The four-column coding structure includes: the first column representing the processing order of the batch; the second column representing the batch's location; the third column representing the processing workers for each batch; and the fourth column representing the task allocation sequence within the batch.
[0016] The decoding rules include inputting the batch into the earliest processing position of the Sairu system row by row according to the four-column encoding structure, taking into account the worker occupancy when determining the start processing time of the batch, assigning tasks to the earliest available worker in sequence within each batch according to the task allocation sequence, and randomly selecting a worker to process if there are multiple available workers.
[0017] Furthermore, in one embodiment of the present invention, the step of performing a probability-based search on the first population and an iterative greedy search on the second population includes:
[0018] Optimizing the worker-batch allocation using a probabilistic model-based search method for the first population includes: establishing a B-row, W-column probability matrix P to describe the distribution information of candidate solutions in the search space, where each element p in P... b,w (b = 1, ..., B; w = 1, ..., W) represents the probability that worker w processes batch b. For batch b, by iterating through the W elements in the b-th row of P, the probability that worker w processes batch b is p. b,w The probability of not processing a batch is 1-p. b,w The workers are selected sequentially to determine whether to process batch b, and finally the workers involved in processing each batch are determined.
[0019] Furthermore, in one embodiment of the present invention, the step of performing a probability-based search on the first population and an iterative greedy search on the second population further includes:
[0020] For the second group, perform iterative greedy search-based optimization of batch assignment and batch sorting, including: in each iteration, optimizing the batch assignment sequence ω and the batch sorting sequence ω. Perform the following operations: deconstruction, reconstruction, and local search; where,
[0021] In the deconstruction phase, Randomly select a position and retrieve all batches after the current position;
[0022] During the reconstruction phase, each batch that was taken out is placed in the position that minimizes the total delay, and the above operation is repeated until all batches are put back.
[0023] During the local search phase, four preset neighborhood search operations are executed sequentially. If an operation improves the target value, the search is re-examined based on the new solution; otherwise, the next operation is executed until all operations are completed.
[0024] Furthermore, in one embodiment of the present invention, the worker-task allocation is optimized through the Q-learning-based local search, wherein the following steps are performed for each batch b:
[0025] Step 1: Set d = 1; count = 0;
[0026] Step 2: Randomly select d consecutive task sequences from the task assignment sequence v. And remove it from v;
[0027] Step 3: Put Insert into all feasible positions in v to form a new task sequence v′. If the target value TT(v′) < TT(v), then replace v with v′ and let count = count + 1.
[0028] Step 4: If count > 0, then let d = 1; otherwise, let d = d + 1.
[0029] Step 5: Determine if the condition d≤d is satisfied. max If the condition is met, proceed to step 2; otherwise, stop the iteration; where d max Indicates the depth of the local search.
[0030] To achieve the above objectives, another aspect of the present invention proposes a scheduling device for a worker transfer system, comprising:
[0031] The population initialization module is used to obtain the coupled subproblems of the Cyrus system scheduling problem, and to construct a first population and a second population for optimizing the coupled subproblems based on preset encoding and decoding rules.
[0032] The dual-population collaborative search module is used to perform a probability-based search on the first population and an iterative greedy search on the second population, and to select the population to enter based on the first search result and the Q-learning selection mechanism.
[0033] The local search module is used to perform a Q-learning-based local search on the second or more individuals in the selected population to obtain a second search result;
[0034] A probability model update module is used to update the probability model based on the second search result and using the second plurality of individuals;
[0035] The termination criterion judgment module is used to determine whether the updated probability model meets the termination criterion. If it does, the result is output.
[0036] The SERROR system scheduling method and apparatus based on worker transfer scenarios in this invention have low computational cost, can not only achieve efficient problem solving and ensure solution efficiency and quality, but also achieve efficient and intelligent scheduling.
[0037] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0038] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:
[0039] Figure 1 This is a flowchart of a SERROR system scheduling method based on a worker transfer scenario according to an embodiment of the present invention;
[0040] Figure 2 This is a schematic diagram illustrating an encoding / decoding example involving five workers and four batches according to an embodiment of the present invention;
[0041] Figure 3 This is a schematic diagram of the structure of the Sairu system scheduling device based on a worker transfer scenario according to an embodiment of the present invention. Detailed Implementation
[0042] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.
[0043] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0044] The following description, with reference to the accompanying drawings, describes a SERROR system scheduling method and apparatus based on a worker transfer scenario, according to an embodiment of the present invention.
[0045] Figure 1 This is a flowchart of the Sairu system scheduling method based on worker transfer scenarios according to an embodiment of the present invention.
[0046] like Figure 1 As shown, the method includes, but is not limited to, the following steps:
[0047] S1, obtain the coupled subproblems of the Selyu system scheduling problem, and based on the preset encoding and decoding rules, obtain a first plurality of individuals to construct a first population and a second population for optimizing the coupled subproblems;
[0048] S2, perform a probability-based search on the first population and an iterative greedy search on the second population, and select the population to enter based on the first search result and the Q-learning selection mechanism;
[0049] S3, perform a Q-learning-based local search on the second plurality of individuals in the selected population to obtain a second search result;
[0050] S4, update the probability model based on the second search result and using the second plurality of individuals;
[0051] S5 determines whether the updated probability model meets the termination criterion. If it does, output the result.
[0052] It is understood that the SSSP-WT considered in this invention is described as follows: W workers with different skill levels process B batches of different types. Each batch contains Q bThere are multiple products of the same type, each requiring different processing tasks. Workers are multi-skilled, meaning they can handle one or more tasks. This problem can be decomposed into four coupled subproblems: 1) Worker-Batch Assignment: Determine the workers for each batch; 2) Worker-Task Assignment: Determine the processing tasks for each worker in a batch; 3) Batch-Cycle Assignment: Determine the batches assigned to each cylinder; 4) Batch Sequencing: Determine the processing order of batches within each cylinder. The problem assumptions are as follows: products are processed in batches, and batches are indivisible; the requirements, type, and standard processing time for each batch are known; workers are allowed to move between cylinders, and movement time is negligible; a task can only be handled by one worker, and each worker can handle multiple tasks. The objective is to minimize the total delay for all batches.
[0053] Specifically, in response to the problem description, the steps of the SERUS system scheduling method based on worker transfer scenarios of the present invention may include:
[0054] Step S1: Population initialization, generating PS individuals;
[0055] Step S2: Establish two populations to optimize different subproblems. Perform a probabilistic learning-based search on population 1 and an iterative greedy search on population 2. Design a Q-learning-based selection mechanism to select the populations.
[0056] Step S3: Select EN = η*PS elite individuals from the population and perform a local search based on Q-learning;
[0057] Step S4: Update the probability model using elite individuals;
[0058] Step S5: Determine whether the termination criterion is met. If it is met, output the result; otherwise, go to step 2.
[0059] The termination criterion is CPU running time: β × W × B.
[0060] Furthermore, the encoding / decoding rules and the design of each stage are elaborated in detail:
[0061] In one embodiment, SSSP-WT needs to consider four coupled sub-problems: worker-batch allocation, worker-task allocation, batch-segment allocation, and batch sorting. A four-column encoding structure is designed to represent a complete solution: the first column represents the processing order of the batch; the second column represents the segment of the batch; the third column represents the processing workers for each batch; and the fourth column represents the task allocation sequence within the batch. The decoding rule is as follows: First, according to the encoding structure, batches are placed row by row into the earliest processing position in the segment system, while considering worker occupancy when determining the start processing time of a batch. Then, within each batch, tasks are assigned sequentially to the earliest available worker according to the task allocation sequence; if multiple available workers exist, a worker is randomly selected for processing. Figure 2 This is an encoding / decoding instance with 5 workers and 4 batches. Following the encoding rules above, PS individuals are randomly generated to form the initial population.
[0062] In one embodiment, the dual-population collaborative search includes probabilistic learning-based search and iterative greedy search, as well as a Q-learning-based selection mechanism.
[0063] Probabilistic Learning-Based Search: For population 1, a probabilistic model-based search method is designed to optimize worker-batch assignment. A B-row, W-column probability matrix P is constructed to describe the distribution information of candidate solutions in the search space. Each element p in P... b,w (b = 1, ..., B; w = 1, ..., W) represents the probability that worker w processes batch b. For batch b, by iterating through the W elements in the b-th row of P, the probability that worker w processes batch b is p. b,w The probability of not processing a batch is 1-p. b,w The workers are selected sequentially to determine whether to process batch b, and finally the workers involved in processing each batch are determined.
[0064] Iterative Greedy Search: For population 2, an iterative greedy search is designed to optimize batch assignment and sorting. In each iteration, the batch assignment sequence ω and the batch sorting sequence are calculated. Perform the following operations: deconstruction, reconstruction, and local search. During the deconstruction phase, in In the initial stage, a random position is selected, and all batches after that position are retrieved. During the reconstruction stage, each retrieved batch is placed in the position that minimizes the total delay. This process is repeated until all batches are returned. In the local search stage, the following four neighborhood search operations are executed sequentially. If an operation improves the target value, the search is re-examined based on the new solution; otherwise, the next operation is executed until all operations are completed.
[0065] Operation 1: From Randomly select two batches and swap their positions;
[0066] Operation 2: From Randomly select one batch and insert it. Any position in the middle;
[0067] Operation 3: From Randomly select two batch positions and reverse the order of all batches in between;
[0068] Operation 4: Randomly select one position in ω and change the batch number of that position;
[0069] A Q-learning-based selection mechanism: Reinforcement learning (RL) technology learns decision-making strategies through continuous interaction between the agent and the environment, maximizing the cumulative reward of the actions performed. RL technology can be used to mine knowledge generated during the search process and guide search behavior, improving search efficiency. This invention integrates the Q-learning algorithm in RL and designs a Q-learning-based selection mechanism to guide individuals in choosing which population to join in each generation. Specific design details are as follows:
[0070] State: Extract the characteristics of different problems as their states. The worker-batch allocation problem is described using the average of all batch processing times after standardization, as shown in Equation (1); the state of the batch allocation and sorting problem is described using the proportion of the number of batches that cause delays to the total number of batches, as shown in Equation (2).
[0071]
[0072] S2=N tardiness / B (2)
[0073] Among them PT min and PT max These represent the minimum and maximum batch processing times (PT) among all batches. b N represents the processing time for batch b. tardiness This represents the total number of batches that experienced delays. It maps the two-dimensional continuous state (S1, S2) to a discrete state. First, the two intervals [0.1] are discretized into F uniformly distributed subintervals. Then, the two subintervals f1 and f2 to which S1 and S2 belong are determined respectively, for example: S1∈[(f1-1) / F, f1 / F], f1∈{1, 2, ..., F}. Finally, the discrete state v of the mapping is determined. f , where f = (f1-1)*F + f2.
[0074] Action: In each generation, the agent perceives the current state and then performs the corresponding action. In this mechanism, two actions are defined as: 1) entering population 1; 2) entering population 2.
[0075] Rewards: The agent maximizes its cumulative gains by performing appropriate actions, so a suitable reward design helps the agent learn efficiently. In RL-CCA, the degree of improvement of the objective function is defined as the reward, as shown in equation (3).
[0076] reward = max(TT) (g-1) -TT (g) / TT (g-1) ,0) (3)
[0077] Among them, TT (g) The total delay for the g-th generation batch is 0 if there is no improvement.
[0078] In one embodiment, Q-learning-based local search is used to enhance the algorithm's local exploration capability. A Q-learning-based local search operation is designed to optimize worker-task allocation, wherein the following steps are performed for each batch b:
[0079] Step 1: Set d = 1; count = 0;
[0080] Step 2: Randomly select d consecutive task sequences from the task assignment sequence v. And remove it from v;
[0081] Step 3: Put Insert into all feasible positions in v to form a new task sequence v′. If the target value TT(v′) < TT(v), then replace v with v′ and let count = count + 1.
[0082] Step 4: If count > 0, then let d = 1; otherwise, let d = d + 1.
[0083] Step 5: Determine if the condition d≤d is satisfied. max If the condition is met, proceed to step 2; otherwise, stop the iteration; where d max Indicates the depth of the local search.
[0084] Using Q-learning techniques, d is continuously adjusted during the search process. max This allows for adjustments to the degree of local search at different stages of the algorithm, improving its search efficiency. Specific design details are as follows: State definitions: 1) S = 1, indicating that the local search operation improves the objective function; 2) S = 0, indicating that the local search operation does not improve the objective function; Action definitions: d max =1,d max =2,d max =3; the reward is defined as shown in equation (3).
[0085] In one embodiment, the probability model is updated. To ensure that the probability model can represent the distribution of the dominant group, EN elite solutions are selected in each iteration to update the probability model. The update formula is as follows:
[0086]
[0087]
[0088] Where λ is the learning rate. The indicator function for the elite solution.
[0089] According to the SPS scheduling method based on worker transfer scenarios in this invention, it can efficiently solve the problem, obtain the optimal solution and achieve intelligent and efficient scheduling. It can also better leverage the flexibility of SPS, improve production efficiency, quickly meet the personalized needs of consumers, and help enterprises enhance their core competitiveness.
[0090] To achieve the above embodiments, such as Figure 3 As shown, this embodiment also provides a SERUS system scheduling device 10 based on worker transfer scenarios. The device 10 includes a population initialization module 100, a dual-population cooperative search module 200, a local search module 300, a probability model update module 400, and a termination criterion judgment module 500.
[0091] Population initialization module 100 is used to obtain coupled subproblems of the Selyu system scheduling problem, and to obtain a first plurality of individuals based on preset encoding and decoding rules to construct a first population and a second population for optimizing the coupled subproblems;
[0092] The dual-population collaborative search module 200 is used to perform a probability-based search on the first population and an iterative greedy search on the second population, and selects the population to enter based on the first search result and the Q-learning selection mechanism.
[0093] The local search module 300 is used to perform a Q-learning-based local search on the second or more individuals in the selected population to obtain the second search result;
[0094] The probability model update module 400 is used to update the probability model based on the second search result and using the second plurality of individuals;
[0095] The termination criterion judgment module 500 is used to determine whether the updated probability model meets the termination criterion. If it does, the result is output.
[0096] Furthermore, the termination criterion is CPU runtime; coupled subproblems include worker-batch allocation, worker-task allocation, batch-Syrian allocation, and batch sorting; the encoding / decoding rules include a four-column encoding structure and decoding rules; wherein,
[0097] The four-column coding structure includes: the first column representing the processing order of the batch; the second column representing the batch's location; the third column representing the processing workers for each batch; and the fourth column representing the task allocation sequence within the batch.
[0098] The decoding rules include inputting the batch into the earliest processing position of the Sairu system row by row according to the four-column encoding structure, taking into account the worker occupancy when determining the start processing time of the batch, assigning tasks to the earliest available worker in sequence within each batch according to the task allocation sequence, and randomly selecting a worker to process if there are multiple available workers.
[0099] Furthermore, the aforementioned dual-population cooperative search module 200 is also used for:
[0100] The worker-batch assignment is optimized using a probabilistic model-based search method for the first group, including: constructing a B-row, W-column probability matrix P to describe the distribution information of candidate solutions in the search space, where each element p in P... b,w (b = 1, ..., B; w = 1, ..., W) represents the probability that worker w processes batch b. For batch b, by iterating through the W elements in the b-th row of P, the probability that worker w processes batch b is p. b,w The probability of not processing a batch is 1-p. b,w The workers are selected sequentially to determine whether to process batch b, and finally the workers involved in processing each batch are determined.
[0101] Furthermore, the aforementioned dual-population cooperative search module 200 is also used for:
[0102] For the second group, perform iterative greedy search-based optimization of batch assignment and batch sorting, including: in each iteration, optimizing the batch assignment sequence ω and the batch sorting sequence ω. Perform the following operations: deconstruction, reconstruction, and local search; where,
[0103] In the deconstruction phase, Randomly select a position and retrieve all batches after the current position;
[0104] During the reconstruction phase, each batch that was taken out is placed in the position that minimizes the total delay, and the above operation is repeated until all batches are put back.
[0105] During the local search phase, four preset neighborhood search operations are executed sequentially. If an operation improves the target value, the search is re-examined based on the new solution; otherwise, the next operation is executed until all operations are completed.
[0106] Furthermore, the aforementioned local search module 300 is also used to optimize the worker-task allocation through the Q-learning-based local search, wherein the following sub-module is executed for each batch b:
[0107] The first submodule is used to set d = 1 and count = 0;
[0108] The second submodule is used to randomly select d consecutive task sequences from the task assignment sequence v. And remove it from v;
[0109] The third submodule is used to... Insert into all feasible positions in v to form a new task sequence v′. If the target value TT(v′) < TT(v), then replace v with v′ and let count = count + 1.
[0110] The fourth submodule is used to set d = 1 if count > 0; otherwise, d = d + 1.
[0111] The fifth submodule is used to determine whether the condition d≤d is satisfied. max If the condition is met, proceed to the second submodule; otherwise, stop the iteration. Where d max Indicates the depth of the local search.
[0112] According to the embodiments of the present invention, the SPS system scheduling device based on worker transfer scenarios can efficiently solve problems, obtain the optimal solution to the problem and realize intelligent and efficient scheduling. It can also better leverage the flexibility of SPS, improve production efficiency, quickly meet the personalized needs of consumers, and help enterprises enhance their core competitiveness.
[0113] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.
[0114] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.
Claims
1. A scheduling method for a SERROR system based on worker transfer scenarios, characterized in that, Includes the following steps: The coupled subproblems of the Selju system scheduling problem are obtained, and multiple individuals are obtained based on preset encoding and decoding rules to construct a first population and a second population for optimizing the coupled subproblems. A probabilistic learning-based search is performed on the first population and an iterative greedy search is performed on the second population. The population to be entered is selected based on a Q-learning selection mechanism. Specifically, a probabilistic model-based search method is used to optimize worker-batch allocation for the first population, and an iterative greedy optimization of batch-Selwyn allocation and batch sorting is performed on the second population. Specifically, the Q-learning selection mechanism is as follows: the state is a two-dimensional state consisting of the standardized mean of all batch processing times and the proportion of batches with delays to the total number of batches; the action is choosing between the first group and the second group to enter; and the reward function is the relative improvement rate. ,in for The total delay in batches of the project; A Q-learning-based local search is performed on multiple elite individuals in the selected population to be entered. The worker-task allocation is optimized through the Q-learning-based local search to obtain a second search result. In the Q-learning-based local search, the depth of the local search... Q-learning adaptively adjusts the state of the objective function based on whether the current local search operation improves it; the action is selection. , or The reward is the relative improvement rate of the objective function; Based on the second search result, the probability model is updated using the multiple elite individuals; Determine whether the updated probability model meets the termination criterion. If it does, output the result.
2. The method according to claim 1, characterized in that, The termination criterion is CPU runtime; the coupled subproblems include worker-batch allocation, worker-task allocation, batch-Syrian allocation, and batch sorting; the encoding / decoding rules include a four-column encoding structure and decoding rules; wherein, The four-column coding structure includes: the first column representing the processing order of the batch; the second column representing the batch's location; the third column representing the processing workers for each batch; and the fourth column representing the task allocation sequence within the batch. The decoding rules include inputting the batch into the earliest processing position of the Sairu system row by row according to the four-column encoding structure, taking into account the worker occupancy when determining the start processing time of the batch, assigning tasks to the earliest available worker in sequence within each batch according to the task allocation sequence, and randomly selecting a worker to process if there are multiple available workers.
3. The method according to claim 2, characterized in that, The step of performing a probability-based search on the first population and an iterative greedy search on the second population includes: Optimizing the worker-batch allocation for the first population using a probabilistic model-based search method includes: establishing a... OK Column probability matrix , is used to describe the distribution information of candidate solutions in the search space, where, P Each element Indicates workers Processing batch The probability for a batch traversal P The Middle b OK One element, worker Processing batch b The probability is The probability of not processing a batch is 1- Select whether the worker processes a batch in turn. Finally, the workers involved in the processing of each batch were determined.
4. The method according to claim 3, characterized in that, The step of performing a probability-based search on the first population and an iterative greedy search on the second population further includes: The second population is optimized for batch-Syrian assignment and batch sorting based on iterative greedy search, including: in each iteration, the batch assignment sequence is optimized. and batch sorting sequence Perform the following operations: deconstruction, reconstruction, and local search; where, In the deconstruction phase, Randomly select a position and retrieve all batches after the current position; During the reconstruction phase, each batch that was taken out is placed in the position that minimizes the total delay, and the above operation is repeated until all batches are put back. During the local search phase, four preset neighborhood search operations are executed sequentially. If an operation improves the target value, the search is re-examined based on the new solution; otherwise, the next operation is executed until all operations are completed.
5. The method according to claim 2, characterized in that, The worker-task allocation is optimized through Q-learning-based local search, wherein for each batch... b Perform the following steps: Step 1: Let count=0; Step 2: From the task assignment sequence Random selection A continuous sequence of tasks and from Remove from; Step 3: Put Insert into All feasible positions are used to form a new task sequence. If the target value Then use Alternative Meanwhile, let count = count + 1; Step 4: If count > 0, then let ;otherwise, ; Step 5: Determine if the conditions are met. If the condition is met, proceed to step 2; otherwise, stop the iteration. Indicates the depth of the local search.
6. A scheduling device for a semaphore system based on worker transfer scenarios, characterized in that, include: The population initialization module is used to obtain the coupled subproblems of the Cyrus system scheduling problem, and to construct a first population and a second population for optimizing the coupled subproblems based on preset encoding and decoding rules. A dual-population collaborative search module is used to perform a probabilistic learning-based search on the first population and an iterative greedy search on the second population, and to select the population to enter based on a Q-learning selection mechanism. Specifically, the first population is optimized for worker-batch allocation using a probabilistic model-based search method, while the second population is optimized for batch-segment allocation and batch sorting using an iterative greedy optimization mechanism. The Q-learning selection mechanism is as follows: the state is a two-dimensional state composed of the standardized mean of all batch processing times and the proportion of delayed batches to the total number of batches; the action is to select either the first or second population to enter; and the reward function is the relative improvement rate. ,in for The total delay in batches of the project; The local search module is used to perform a Q-learning-based local search on multiple elite individuals in the selected population, and optimize worker-task allocation through the Q-learning-based local search to obtain a second search result; In the Q-learning-based local search, the depth of the local search... Q-learning adaptively adjusts the state of the objective function based on whether the current local search operation improves it; the action is selection. , or The reward is the relative improvement rate of the objective function; The probability model update module is used to update the probability model for the multiple elite individuals. The termination criterion judgment module is used to determine whether the updated probability model meets the termination criterion. If it does, the result is output.
7. The apparatus according to claim 6, characterized in that, The termination criterion is CPU runtime; the coupled subproblems include worker-batch allocation, worker-task allocation, batch-Syrian allocation, and batch sorting; the encoding / decoding rules include a four-column encoding structure and decoding rules; wherein, The four-column coding structure includes: the first column representing the processing order of the batch; the second column representing the batch's location; the third column representing the processing workers for each batch; and the fourth column representing the task allocation sequence within the batch. The decoding rules include inputting the batch into the earliest processing position of the Sairu system row by row according to the four-column encoding structure, taking into account the worker occupancy when determining the start processing time of the batch, assigning tasks to the earliest available worker in sequence within each batch according to the task allocation sequence, and randomly selecting a worker to process if there are multiple available workers.
8. The apparatus according to claim 7, characterized in that, The dual-population cooperative search module is also used for: Optimizing the worker-batch allocation for the first population using a probabilistic model-based search method includes: establishing a... OK Column probability matrix , is used to describe the distribution information of candidate solutions in the search space, where, P Each element Indicates workers Processing batch The probability for a batch traversal P The Middle b OK One element, worker Processing batch b The probability is The probability of not processing a batch is 1- Select whether the worker processes a batch in turn. Finally, the workers involved in the processing of each batch were determined.
9. The apparatus according to claim 8, characterized in that, The dual-population cooperative search module is also used for: The second population is optimized for batch-Syrian assignment and batch sorting based on iterative greedy search, including: in each iteration, the batch assignment sequence is optimized. and batch sorting sequence Perform the following operations: deconstruction, reconstruction, and local search; where, In the deconstruction phase, Randomly select a position and retrieve all batches after the current position; During the reconstruction phase, each batch that was taken out is placed in the position that minimizes the total delay, and the above operation is repeated until all batches are put back. During the local search phase, four preset neighborhood search operations are executed sequentially. If an operation improves the target value, the search is re-examined based on the new solution; otherwise, the next operation is executed until all operations are completed.
10. The apparatus according to claim 7, characterized in that, The local search module is further configured to optimize the worker-task allocation using the Q-learning-based local search, wherein for each batch... b Execute the following submodules: The first submodule is used to enable... count=0; The second submodule is used to allocate sequences from tasks. Random selection A continuous sequence of tasks and from Remove from; The third submodule is used to... Insert into All feasible positions are used to form a new task sequence. If the target value Then use Alternative Meanwhile, let count = count + 1; The fourth submodule is used to let if count > 0. ;otherwise, ; The fifth submodule is used to determine whether the conditions are met. If the condition is met, proceed to the second submodule; otherwise, stop the iteration. Indicates the depth of the local search.
