Research On Hyper-heuristic Algorithm For Solving Flexible Job Shop Scheduling Problem

The flexible job shop scheduling problem is a classic and important optimization problem in production scheduling.Research on this problem improves enterprise production management efficiency,promotes manufacturing automation and intelligence,and supports industrial upgrading.Current optimization methods mainly rely on heuristic and meta-heuristic algorithms,which have strong performance and high efficiency but also have limitations.Hyper-heuristic algorithm frameworks have emerged as a new approach to optimize complex operations research problems.These frameworks separate high-level strategies from low-level problem domains,making them easy to expand,with low coupling and strong universality.Under the guidance of the hyper-heuristic framework,new feasible methods for solving flexible job shop scheduling problems can be designed.In the process,different methods can be introduced as high-level strategies for hyper-heuristics,exploring the roles and collaboration modes of deep reinforcement learning and metaheuristic methods in the development and design of hyper-heuristic algorithms.Furthermore,research can be done on the application methods and key elements of performance improvement of hyper-heuristic algorithm frameworks in this field.Based on the idea of selecting hyper-heuristic algorithms,and combining knowledge in the flexible job shop scheduling field,a series of low-level heuristic methods can be designed according to the selected encoding method for framework invocation and operation.A reinforcement learning hyper-heuristic algorithm can be obtained by using deep reinforcement learning as a high-level selection strategy based on this framework.By further analyzing the characteristics and problems displayed by the algorithm above in actual operation and understanding the operating mode,it can be improved by designing a hybrid strategy-based automatic design hyper-heuristic algorithm.This algorithm reuses most of the low-level heuristic methods,with an iterative local search process as the algorithm design template,using a variable neighborhood descent process as the local search method,and introducing a deep reinforcement learning agent for algorithm parameter design decisions,thus achieving stronger optimization capabilities.After the algorithm design is completed,experiments are carried out on the two algorithms and the results are analyzed.First,algorithm parameter experiments are performed to determine the ideal parameter settings for algorithm operation.Subsequently,performance comparison experiments are conducted,comparing the two algorithms with heuristic algorithms based on scheduling rules,various classic meta-heuristic algorithms,and direct optimization deep reinforcement learning algorithms.The overall mechanism and the effects of key sub-strategies of the algorithms are also experimentally validated and analyzed.Finally,a comparative analysis of the performance and stability of the two designed algorithms is conducted.Experimental results prove that the proposed hyper-heuristic algorithm effectively optimizes the flexible job shop scheduling problem,with impressive optimization performance,universality,and expandability.Selecting excellent neighborhood structures,effectively organizing the operation of low-level heuristic methods,and enhancing the ability to escape local optima are keys to improving the optimization performance of hyper-heuristic algorithms.Applying deep reinforcement learning strategies can play an auxiliary role in improving optimization performance.