Multivariate Predictive Control Strategy for Multi-Rate Systems Based on Neural Network

In this paper, the control method of multi-rate system based on dynamic Matrix control (DMC) strategy and neural network interpolation technique is studied. Aiming at the problem that the sampling period of the slow sampling output channel is not synchronized with the reference period in the multi-rate system, an improved model predictive control strategy is proposed. The proposed method improves the accuracy of the DMC control model by predicting the missing step response values through neural networks. At the same time, the prediction error filter is designed to cope with the specificity of the multi-rate system and enhance the system stability. Compared with traditional methods such as quadratic spline interpolation, the proposed method effectively solves the prediction and control problems of inconsistent sampling rates and non-sampling points, and improves the applicability in complex industrial processes. Finally, the proposed control strategy was applied to the simulation system of a typical heavy oil fractionating column. By setting different slow sampling periods and model mismatch degrees, it is verified that the DMC control algorithm based on neural network is significantly superior to the DMC control algorithm based on quadratic spline interpolation in terms of convergence speed and stability.