Iterative b-spline neural networks for stochastic distribution control and its application in industrial process

Iterative learning of B-spline basis functions model for the output probability density function (PDF) control of non-Gaussian systems is studied in this paper using the recursive least square algorithm. Within each control interval, the basis functions are fixed and the control input design is performed that controls the shape of the output PDFs. However, between each control interval, periodic learning techniques are used to tune the shape of the basis functions. This has been shown to be able to improve the accuracy of the B-spline approximation model. As such, the overall B-spline model of the output PDFs becomes a dual-model related to both time and space variables. The algorithm has been applied to a simulation study of the molecular weight distribution (MWD) control of a styrene polymerization process, leading to some interesting results.