多维度柔性传送带给料加工站的优化控制模型与方法及鲁棒性

Firstly, according to the physical structure and control mode of the multidimensional flexible conveyor-serviced production station (CSPS), we study its working mechanism and stochastic dynamic characteristics. In the mode of multi-variety mixed-flow production and demand-driven production, we will establish the mathematical model for optimal control and present the corresponding theoretical optimization algorithm, especially the theoretical optimization method based on soft state aggregation when picking position, look-ahead range and service rate of the system are controllable. Meanwhile, optimal settings of physical parameters will also be studied. Secondly, the optimal robust control model and solution of corresponding optimal policy are investigated under the cases where arrival rates of both parts and demands, and service rate of parts are uncertain. Thirdly, we study the online data-driven optimal algorithms, focusing on neuron-dynamic programming methods by combining soft state aggregation and approximate network structure. Finally, semi-Markov modulated process is utilized to consider the robustness of the optimal policy obtained from the theoretical algorithm and the online optimization algorithm as well as its stopping policy. The results will reveal the mapping rules from multidimensional state space to multidimensional action space, clarify the optimal control mechanism of system, and provide some approaches and references for the optimal design and control of a real CSPS system. Moreover, the research will also be a beneficial supplement for the optimal control of sophisticated stochastic decision systems, and lay the foundation for the study of multi-station and multidimensional flexible CSPS.

首先，根据多维度柔性传送带给料加工站（CSPS）的物理结构和控制方式，研究其工作机理和随机动态特性，建立多品种混流生产和需求驱动生产模式下系统捡取点、前视距离和服务率三者可控的优化控制数学模型及其理论优化算法，重点研究基于状态软聚类的理论优化方法，并研究系统物理配置参数的最优设计方法；其次，研究系统工件到达率、需求到达率或服务率不确定情况下的最优鲁棒控制数学模型和最优鲁棒控制策略求解方法；然后，研究系统的数据驱动在线优化算法，重点研究基于状态软聚类和网络逼近结构的神经元动态规划法；最后，运用半马尔可夫调制过程，研究系统的最优策略、在线优化算法及其停止策略的鲁棒性能评价方法。研究成果将揭示系统多维状态空间到多维行动空间的最优映射规律，阐明其最优控制机制，为实际CSPS系统的优化设计与控制提供解决思路和参考依据，也是对复杂随机决策系统优化方法的补充，并为多站点多维度柔性CSPS的研究奠定基础。

随着科学技术的快速发展，人类社会将逐步进入“工业4.0”时代，迎来以智能制造为主导的第四次工业革命。而在许多实际制造企业中，存在一类由传送带给料生产加工站(CSPS)构成的现代化生产线。在某种程度上，CSPS的智能化水平反映了整个产线甚至该企业的信息化水平。因此，对这类系统的优化控制问题的研究，具有十分重要的现实意义。本项目研究了多维度柔性CSPS的优化控制问题以及鲁棒性评价方法。首先，建立了多品种混流生产和需求驱动生产模式下多维度柔性CSPS的理论优化模型，研究了理论优化算法以及系统物理配置参数的最优设计方法；其次，研究了系统工件到达率、需求到达率或服务率不确定情况下的最优鲁棒控制数学模型和最优鲁棒控制策略求解方法；然后，研究了基于数据驱动的在线优化控制机制和方法，其中重点研究了神经元动态规划方法；最后，研究了在工件随机到达特性等不为泊松分布时，系统的最优策略、在线优化算法及其停止策略的鲁棒性能评价方法。研究成果是对复杂离散事件动态系统优化方法的一个有益补充，同时也为一类智能制造或服务系统的实际优化控制提供新的思路和方法。