Optimal control of PVC polymerization process with set values based on TD3 algorithm.

Abstract

Aiming at the various time frames between the operation computing and the base loop computing of the PVC polymerization process, as well as the difficulty in establishing the nonlinear model of the operation layer, this paper presents a data-based run-level control method combining iterative lifting technique and the TD3 algorithm. Aiming at the problem of different time scales of the two-layer structure, the iterative boosting technique is used to boost the cycle of the basic level of circulation to the cycle of the operation layer, and substitute the closed-loop basic loop system into the operational layer model; then the operation layer model is augmented and generalized, and general control objects that use the set values of the cyclic layer as input values, and with the operation index as the output value, are obtained. Based on the synchronized updating of the value function and the control strategy, the online strategy iteration algorithm is implemented using a model-free method based on TD3 neural network.