复杂工业过程非串级双速率组合分散运行优化控制

复杂工业过程具有模型维数高、多时间尺度耦合、动态不确定性等特点,其运行优化控制(Operational optimal control, OOC)一直是控制领域的研究难点与热点.本文聚焦一类由多个快变且互联的设备单元与慢变且模型未知的运行过程串联组成的工业过程,提出一种数据和模型混合驱动的非串级双速率组合分散运行优化控制方法.该方法通过奇异摄动理论,将非串级双速率运行优化问题描述为异步采样的慢子系统最优设定值跟踪和快子系统最优调节控制.利用工业运行数据,采用不依赖系统动态的Q-学习算法设计慢子系统最优跟踪策略,克服运行过程模型难以建立的情形;针对快子系统,设计基于模型的分散次优控制策略,并给出收敛因子的下界,解决设备层互联项对系统稳定性的影响.通过浮选过程仿真实验验证了所提控制方法的有效性.

Non-cascade Dual-rate Composite Decentralized Operational Optimal Control for Complex Industrial Processes

Due to the features of high model dimensionality, multiple time scales, and dynamic uncertainty, operational optimal control (OOC) for complex industrial processes has always been a difficult and hot topic in the control community. This paper proposes a data-based and model-based non-cascade dual-rate composite decentralized OOC scheme for a class of industrial processes, which consist of multiple fast-changing and interconnected unite devices and a slow-changing operational process with unknown dynamics. Based on singular perturbation theory, the non-cascade dual-rate OOC problem is formulated as two asynchronous sampling optimal problems related to set-point tracking of the slow subsystem and regulator of the fast subsystem. The Q-learning algorithm which is independent of the model knowledge of the operational process is adopted to design optimal tracking policy for the slow subsystem by using industrial data. For the fast subsystem, we present a model-based decentralized suboptimal control policy and provide a lower bound on the convergence factor, so that the impact of interconnected items on system stability is offset. Simulation experiments on a flotation process illustrate the effectiveness of the proposed method.