An efficient method for parameter estimation of a nonlinear system using Backtracking Search Algorithm

Abstract

Tank level systems are ubiquitous in process industries and exhibit a nonlinear nature. This nonlinear behaviour arises prominently due to nonlinear dependence of outflow rate on tank level and/or due to presence of other nonlinear elements in loop such as nonlinear actuators. It is essential to completely investigate the dynamics of system so as to generate an effective and accurate process model for obvious reasons. It may be noted that it is hard to find a unified model of a nonlinear system which can characterize the process in the entire operating range. To address this issue, this paper is a sincere effort to accurately identify a nonlinear tank level system with the help of a recently developed evolutionary algorithm i.e. Backtracking Search Algorithm (BSA). Two different optimization problems were created and solved using BSA to estimate the tank level system parameters and are termed as Method 1 and Method 2, respectively. The optimization was performed to minimize summation of absolute error (SAE) between experimental and simulated data yielding the parameters of model. During the training phase, Method 1 used only one data set and Method 2 used a combination of five different data sets of experiments at different operating conditions. From the results obtained through the experimentation on a hardware setup, it can be easily inferred that parameter estimation using Method 2 gives better identification than Method 1 by providing lower SAE values.