Modeling error PDF shape based data-driven model for batch processes

Abstract

The key of optimal control of batch processes depends on obtaining an accurate model. The data-driven modeling method based on process input-output data points is a hot spot in the study of batch process modeling. This paper breaks through the idea that mean squared error (MSE) is employed as the index function in traditional data-driven modeling method, and a novel data-driven modeling approach for batch process is proposed. The conception of probability density function (PDF) control is firstly introduced and then a model error control system of batch process is built. The adjustable parameters of the data-driven model are taken as the control system input, and the shape of the PDF as the corresponding output. As a result, the open-loop model parameter identification problem is transferred into a closed-loop control problem of the shape of PDF. The adjustable parameters dominate the spatial distribution of the PDF of the model error, which not only guarantees the accuracy of the model but also eliminates the colored noise. Simulation results demonstrate that the proposed data-driven modeling approach based on the shape of PDF has better modeling precision, robustness and generalization ability; and also provides a new way for the data-driven modeling of batch processes.