基于强化学习的分布参数系统数据驱动控制

Most of practical industrial processes are distributed parameter system (DPS), which are essentially described by a set of complex nonlinear partial differential equations. Due to the infinite-dimensional nature of the DPSs, the direct application of the control theories and methods for lumped parameter systems to them is impossible. Moreover, with the fast developments of science technologies, many industrial processes become more and more complicated due to their large scale and complex manufacturing techniques, equipments and procedures. Therefore, the accurately modeling and identification of these processes are often costly to conduct, or the established models are too complicated to support controller design. To overcome these difficulties, this project aims at studying reinforcement learning methods for data-driven control problem of the DPSs, and establishing its theories for performance and stability analysis. The effectiveness and the practical feasibility of the methods will be overified with computer simulations. Through the research of this project, some novel and effective methods and theories will be provided for control design of DPSs, which are extremely important for the development of data-driven control theories and meaningful in both scientific researches and real engineering applications.

大部分实际工业过程均为分布参数系统(DPS)，它们本质上由复杂的非线性偏微分方程描述，DPS具有无穷维自由度的特征，所以现有针对集中参数系统的控制理论与方法无法直接用于DPS。而且，随着科学技术的快速发展，工业系统的规模越来越大，生产过程越来越复杂，导致精确建立DPS数学模型的代价非常大，或是模型非常复杂而无法用于控制器设计。为解决这一困难，本项目拟引入强化学习的思想，研究DPS数据驱动控制问题，建立相应的性能分析与稳定性理论，并通过计算机仿真，验证方法的有效性，和探讨实际应用的可行性。通过对本项目的研究，将为DPS的控制设计提供一些新的、有效的方法和理论依据，促进数据驱动控制理论的发展，具有重要的科学意义和应用价值。

大部分实际工业过程均为分布参数系统(DPS)，它们本质上由复杂的非线性偏微分方程描述，DPS具有无穷维自由度的特征，所以现有针对集中参数系统的控制理论与方法无法直接用于DPS。而且，随着科学技术的快速发展，工业系统的规模越来越大，生产过程越来越复杂，导致精确建立DPS数学模型的代价非常大，或是模型非常复杂而无法用于控制器设计。为解决这一困难，本项目引入强化学习的思想，研究了数据驱动最优控制问题，提出了一系列基于强化学习的控制方法及理论。相关成果发表了SCI期刊论文16篇，国际学术会议论文3篇，包括领域顶级期刊：IEEE Transactions on Cybernetics, IEEE Transactions on Neural Networks and Learning Systems, IEEE Transactions on Industrial Electronics。