Statistic information tracking of Non-Gaussian systems: A data-driven control framework based on adaptive NN modeling

Abstract

A new type of data-driven control framework for Non-Gaussian stochastic systems is established in this paper. Different from the traditional feedback style, the driven information for tracking problem is the statistic information set (SIS) of the output rather than the output value. The set of statistical information (including the moments and the entropy) or probability density functions (PDFs) of the output are the measured information and the controlled objective. Under this framework, a mixed two-step adaptive neural network (NN) modeling is established with combining a static NN for description of the statistic information or PDFs and a dynamic one for identification of the relationship between input and output weight vectors. An adaptive PI tracking controller based on the proposed dynamic NNs is designed so as to track a target stochastic distribution. Finally, simulation results on a model in paper-making processes are given to demonstrate the effectiveness.