Multi-rate Layered Optimal Operational Control Of Industrial Processes

The connotation of optimal operational control for industrial process is to integrate the knowledge and data of production process using information technology.Through optimal operation and control technology,could not only stabilize the controlled system,ensuring the output of basic loop layer tracking the set-points.But also could control the entire running process,ensuring the operating indicators that characterize the production efficiency,quality and energy consumption of products is optimal.Optimal operational control for industrial process involves operation layer and the basic loop layer,known as two time-scale characteristic,and the layered control is an easy-to-implement engineering choice for it.However,the design of optimal operational controller for industrial process faces complex multi-rate problems because of various instruments obtain information and process information at different speeds due to their own particularities,the control nodes are dispersed,and the control update rate is not consistent with the data sampling rate of the actual industrial process control system.At the same time,it is difficult to establish a mathematical model of the operation layer,meaning the traditional control methods cannot be applied directly.Therefore,this paper we conducts the research on multi-rate layered optimal operational control of industrial processes under the support of the National Natural Science Foundation of China project “Data-Driven Operation Optimization Control of Whole Grinding Process in Dynamic Environment”.A multi-rate layered optimal operational control of industrial process method is proposed in this paper by combing the method of multi-rate problem solution and data-driven reinforcement learning.The main research work is as follows:(1)A solution for industrial processes which has multi-time scale and multi-rate layered structure is proposed.First,raise the input and output sampling period of the basic loop layer to the frame period using the block lifting technology,resulting designs the tracking controller of loop layer;second,unify the time scale of the basic loop layer and the operation layer using the recursive lifting technology,resulting designs the setpoints optimization controller of operation layer.(2)Aiming at a class of industrial processes that the operation layer can be approximated linearly,a data-driven optimal operational control method combining lifting technology,model predictive control and Q-learning is proposed,and the convergence proof of it is given.Comparing with the offline solution method based on Riccati equation and the policy iteration algorithm based on Bellman equation,this method has the advantage of not relying on system dynamic information,using online data only.Finally,this method is applied to the metallurgical grinding process for simulation research.(3)Aiming at a class of industrial processes that the operation layer is an affine nonlinear system,a data-driven optimal operational control method combining lifting technology,model predictive control and actor-critic neural network is proposed,and the convergence proof of it is given.This method has the advantages of not relying on the model of operation layer.Finally,this method is applied to dense medium coal preparation process in coal production for simulation research.