随着社会经济的发展和生产技术水平的提升，制造系统开始向高端化、智能化、绿色化发展，出现柔性、绿色、分布式等新模式，以满足社会生产的多元需求和动态变化。目前，基于学习机制的优化技术在处理复杂调度优化问题时表现出卓越性能，然而在机制层面尚缺乏系统性开发和针对性设计。本学位论文从流水车间到作业车间，聚焦柔性、绿色、分布式和多阶段等新型制造系统，开展生产、运输和装配集成的多层级复杂调度优化研究，形成基于学习的策略自生成机制、算法自适应机制和模态自组织机制，为智能制造生产调度提供理论和方法支撑。论文在问题层面，综述多类典型制造系统调度研究进展；在方法层面，综述基于学习机制的主流调度优化技术发展情况，并深入研究取得如下成果：（1）针对制造系统流水车间调度问题，提出一种深度学习驱动的策略自生成机制，设计符合问题特点的编解码网络，结合强化学习、监督学习和局部搜索，构建知识引导的端到端优化模式，进而提出一种基于深度学习的调度优化算法。（2）针对柔性制造系统能效调度问题，提出一种反馈学习增强的算法自适应机制，结合群体统计信息建立种群质量评估指标，动态选择高效演化操作，反馈调节算法搜索模式，进而提出一种基于反馈学习的调度优化算法。（3）针对柔性制造与运输协同调度问题，提出一种强化学习调节的算法自适应机制，以多种群结构为框架融合多种搜索模式，利用强化学习实现个体的自适应演化，设计知识驱动的局部搜索，进而提出一种基于强化学习的调度优化算法。（4）针对分布式制造与运输协同能效调度问题，提出一种统计学习辅助的模态自组织机制，综合利用统计学习和演化学习的优势，协同使用基于分解和Pareto支配两种多目标处理机制，根据搜索程度自组织调整搜索模态，实现全局探索和局部开发的平衡，进而提出一种基于统计学习的调度优化算法。（5）针对分布式多阶段制造运输集成调度问题，提出一种分层学习引导的模态自组织机制，融合特定点和非支配解集搜索，结合问题特征，分层引导搜索模态的自组织，实现分层-融合协同，进而提出一种基于分层学习的调度优化算法。论文通过大量仿真实验和统计分析，全面测试算法参数灵敏度，充分验证算法策略有效性。通过与代表性算法对比，验证基于学习机制的调度优化算法的优越性。

With the advancement of the national economy and the improvement of production technology, manufacturing systems are evolving towards high-end, intelligent, and environmentally friendly directions. Emerging trends such as flexibility, sustainability, and distribution are addressing the diverse and dynamic demands of social production. At present, optimization methods based on learning mechanisms show remarkable performance in solving complex scheduling optimization problems. However, further research is needed in the aspect of mechanisms. Centered on traits such as flexibility, sustainability, distribution, and multi-staging, this dissertation introduces the learning-based policy self-generation mechanism, algorithm adaptation mechanism, and modal self-organization mechanism for intelligent manufacturing production scheduling.This dissertation reviews research on various typical intelligent manufacturing systems and summarizes the state-of-the-art learning-based scheduling methods. Subsequently, through extensive research, the following results have been achieved.(1) For the flowshop scheduling in intelligent manufacturing, a policy self-generation mechanism based on deep learning is proposed. First, a novel policy network considering the problem characteristics is designed. Then, the knowledge-guided end-to-end optimization mode is constructed by combining reinforcement learning, supervised learning, and local search. Finally, a deep learning-based optimization algorithm is proposed.(2) For the energy-efficient scheduling in flexible manufacturing, an adaptation mechanism based on feedback learning is proposed. An effective method is presented to evaluate the quality of two populations and a feedback mechanism based on population quality is adopted to dynamically adjust the size of each population. A novel process of reproduction, crossover and mutation is developed based on feedback. Finally, the feedback learning-based optimization algorithm is proposed.(3) For the production and transportation scheduling in flexible manufacturing, an adaptation mechanism based on reinforcement learning is proposed. First, the multipopulation strategy is introduced to integrate various search methods. Second, a reinforcement learning-based mating selection is proposed to realize the cooperation of different subpopulations by selecting appropriate individuals for evolutionary search. Then, a speci?c local search inspired by the problem properties is designed to enhance the exploitation capability of the algorithm. Finally, the reinforcement learning-based optimization algorithm is proposed.(4) For the production and transportation scheduling in distributed energy-efficient flexible manufacturing, a mode self-organization mechanism based on statistical learning is proposed. First, statistical learning and evolutionary learning are integrated into the framework, while decomposition and Pareto dominance methods are employed in different stages to handle conflicting objectives. Then, according to the performance of statistical learning, a novel switching mechanism between statistical learning and evolutionary learning is designed to ensure the rational allocation of computing resources. Finally, the statistical learning-based optimization algorithm is proposed.(5) For the integrated scheduling of production, transportation and assembly in distributed multi-stage manufacturing, a mode self-organization mechanism based on hierarchical learning is proposed. Single-objective search and multi-objective search are integrated to establish a bi-hierarchical optimization model. Then, the knowledge-based local searches are designed to strengthen the exploitation of the method by leveraging the interconnections among subproblems. Meanwhile, considering the characteristics of the problem, individual cooperation is employed to enhance the search efficiency. Finally, the hierarchical learning-based optimization algorithm is proposed.The dissertation conducts extensive simulation experiments and statistical analyses. It comprehensively tests the sensitivity of algorithm parameters and thoroughly verifies the effectiveness of algorithm strategies. Through comparisons with other algorithms, the superiority of learning mechanism-based scheduling optimization algorithms is confirmed.