Intelligent Optimization Algorithms and Machine Learning Methods for Lot-streaming Flowshop Scheduling Problems

Efficient scheduling of production processes is a cornerstone of modern manufacturing, directly impacting operational performance and cost efficiency. In the realm of manufacturing and production, flowshop scheduling plays an essential role in optimizing production scheduling to improve efficiency and reduce idle times. In the traditional flowshop scheduling problem, each job can only proceed to the next machine after completing processing on the current one, potentially leading to idle times for downstream machines and extended overall processing times. However, in practical production scheduling, jobs are often subdivided into smaller sublots that can be processed independently. This approach, known as Lot-Streaming (LS) technology, significantly reduces machine idle times and shortens the makespan, thereby enhancing production efficiency. The Lot-streaming Flowshop Scheduling Problem (LFSP) addresses inefficiencies inherent in traditional flowshop scheduling by dividing jobs into sublots, enabling sequential processing across machines and reducing machine idle times. However, LFSP extends beyond the classical flowshop structure, presenting a complex scheduling challenge as it requires both job sequencing and allocation of production resources. This additional complexity makes LFSP more relevant and beneficial to real-world production but also necessitates advanced optimization strategies to address the increased intricacy. This thesis systematically investigates the LFSP and its various extensions, aiming to address complex industrial production challenges through algorithmic innovations. The study begins with an overview of LFSP, including its problem definition, mathematical formulation, and key characteristics. A comprehensive literature review traces the evolution of LFSP from early exact algorithms to heuristic, metaheuristic, and intelligent optimization algorithms, driven by increasing problem complexity and real-world applicability. The structure of this thesis is illustrated in Figure 1. This thesis presents the application of an Efficient Migrating Birds Optimization Algorithm with Speedup Technique (EMBOST) to solve the Lot-streaming Flowshop Scheduling Problem with Equal-size Sublots (ELFSP). The proposed EMBOST algorithm integrates several advanced mechanisms to enhance optimization performance. Specifically, the Nawaz-Enscore-Ham (NEH) heuristic is adopted to construct a high-quality initial population and maintain solution diversity. To accelerate the search process, two decoding methods and a speedup insertion technique are introduced, significantly reducing computational time. Furthermore, a greedy reconstruction strategy is employed to refine solutions, while mixed neighborhood structures are used to enhance local search efficiency and ensure population stability. To overcome premature convergence, an improved reset mechanism based on the Glover operator is incorporated, allowing more effective exploration of the solution space. Experimental results demonstrate that EMBOST consistently outperforms existing algorithms in terms of solution quality and computational efficiency when applied to ELFSP benchmark instances. These results confirm the algorithm’s scalability, particularly in large-scale problem settings. Overall, it’s not only validates the effectiveness of the EMBOST algorithm in addressing ELFSP but also underscores its potential for broader application in complex scheduling optimization problems in industrial environments. This thesis investigates the Hybrid Flowshop Scheduling Problem with Lot-streaming and Random Breakdowns (RBHLFSP), an extension of the traditional LFSP that incorporates both parallel machines and the dynamic nature of real-world production environments. Unlike conventional LFSP models that assume a single machine per stage and deterministic processing, the RBHLFSP reflects practical settings where multiple parallel machines are available at each stage, and unexpected machine breakdowns may occur. Two distinct breakdown scenarios are considered: one where job processing resumes after a breakdown, and another where interrupted jobs must be reprocessed. To efficiently address these complex conditions, an Improved Migrating Birds Optimization (IMBO) algorithm is proposed. The algorithm incorporates enhancements such as NEH-based initialization, a Shortest Waiting Time (SWT) rule, mixed neighborhood structures, and a reset mechanism to improve performance under uncertainty. Comprehensive simulation experiments are conducted to evaluate the algorithm across both breakdown scenarios. Results demonstrate that the IMBO algorithm consistently achieves lower total flow times and exhibits strong adaptability under dynamic conditions. These findings highlight the algorithm’s capability in handling complex scheduling challenges and its potential for application in real-world manufacturing systems. To address ELFSP from a reinforcement learning perspective, an Improved Dynamic Q-Learning (IDQL) algorithm is proposed to address the ELFSP, a complex variant of the classic flowshop scheduling problem. Building on the reinforcement learning framework, the proposed IDQL algorithm utilizes makespan-based feedback to guide learning and employs a dynamic ε-greedy exploration strategy to balance exploration and utilization, thereby avoiding blind search. To ensure solution diversity and enhance convergence to high-quality solutions, the NEH heuristic algorithm is integrated into the initialization process. Furthermore, the Glover operator is incorporated to refine local search and address the limitations of the ε-greedy strategy by improving search intensity around promising areas. Extensive computational experiments are conducted to benchmark the performance of IDQL against other state-of-the-art intelligent optimization algorithms. The results demonstrate that the proposed algorithm consistently achieves superior performance in minimizing makespan across various problem instances. In multi-objective scenarios, the thesis introduces the Multi-objective Hybrid Flowshop Scheduling Problem with Lot-streaming (MHLFSP), which incorporates multiple stages, parallel machines, and job splitting into sublots, aiming to enhance production scheduling efficiency. Unlike traditional scheduling models, MHLFSP requires simultaneous optimization of conflicting objectives, primarily makespan minimization and machine idle time reduction. To address this complex problem, a Multi-objective Q-Learning (MQL) algorithm is proposed. The approach incorporates two distinct reward functions aligned with the objectives and utilizes three Q tables to store Q values: one for each individual objective and one for the combined outcome. A dynamic ε-greedy exploration strategy is implemented to improve search effectiveness and avoid premature convergence, while a greedy local search algorithm refines solution quality. Simulation experiments benchmark the proposed MQL algorithm against existing methods, results demonstrate that the MQL algorithm achieves a better balance between objectives and generates solutions that closely approach the Pareto front. This confirms its efficiency in handling complex multi-objective scheduling problems. In conclusion, this thesis presents a comprehensive exploration of the LFSP and its various extensions, addressing both deterministic and dynamic environments, as well as single-objective and multi-objective contexts. Through the development and application of advanced algorithms—EMBOST, IMBO, IDQL, and MQL—this thesis demonstrates the effectiveness of hybrid metaheuristics and reinforcement learning techniques in solving complex scheduling problems. The proposed methods not only achieve superior performance across benchmark datasets but also offer strong scalability and adaptability to real-world manufacturing scenarios. By integrating classical heuristics, dynamic exploration strategies, and intelligent local search mechanisms, the algorithms provide robust and efficient solutions. This research contributes both theoretical insights for improving production scheduling efficiency, highlighting the potential of intelligent optimization techniques in addressing the increasing complexity of modern manufacturing systems.