工业过程多速率分层运行优化控制

工业过程运行优化控制通常采用基础回路层和运行层两层结构，涉及不同时间尺度特性的被控对象，且由于检测装置采样周期不同难以统一控制与采样周期；此外，运行层动态往往机理复杂难以建模.因此针对这一多层次、多时间尺度且部分模型未知的复杂多速率控制问题，本文提出一种工业过程多速率分层运行优化控制方法.该方法在使用提升技术解决分层多速率问题的基础上，采用一种基于Q-!学习的数据驱动运行层设定值优化方法，更新基础回路层的设定值；并针对提升后的系统采用模型预测控制（Model predictive control，MPC）方法设计基础回路层控制器以跟踪设定值，从而实现运行指标的优化控制.对典型工业闭路磨矿过程进行了仿真实验，验证了本文所提方法的有效性.

Multi-rate Layered Optimal Operational Control of Industrial Processes

The optimal operational control of industrial processes usually adopts a two-layer structure (i.e., the operation layer and basic loop layer), which involves the controlled objects with different-time-scale characteristics, and has unsynchronized control and the sampling periods because of the difference of the sensor sampling periods. In addition, the dynamics of the operation layer is usually too complex to be modeled. Therefore, in order to solve this complex multi-rate control problem with characteristics of multi-layer, multi-time scale and unknown partial dynamics, this paper proposes a multi-rate layered optimal operational control method for industrial processes. On the basis of using the lifting technology to solve the multi-rate layered problem, a data-driven set point optimization method based on Q-learning is first employed to update the basic loop setpoints; and then a model predictive control is used to design the base loop controller according the lifted system to track the updated setpoints, thereby realizing the optimization control of operation indices. Experiments have been carried out on a closed-loop grinding process, which shows the effectiveness of the proposed method.