Design of Neural Estimator for Non-Linear Interacting Process

Abstract

Real processes often exhibit non-linear behavior; time variance and delay between input and output, having a vast amount of highly correlated data and this correlation need to be utilized in the design of estimator. In this work, an interacting non-linear (conical tank) is taken for study and a neural estimator is designed using Data driven approach. The neural network is developed in hardware on a customized Linux kernel and implemented using python language. The second order training algorithm is modified using a batch update strategy and this approach along with the hardware implementation reduces the estimator training period and makes it highly suited to online.