PLS-Based Multivariate Statistical Approach for Soft Sensor Development in WWTP

Abstract

Multivariate Statistical Process Control is a projection method of projecting a high- dimensional model space with a number of measured variables to a low-dimensional space. The different methods include Principal Component Analysis, Principal Component Regression, Partial Least Squares Regression, Finite Impulse response, Autoregressive exogenous input and autoregressive moving average, etc. The advantage of using Multivariate Statistical Process Control is that it identifies low-dimensional quality process data while reducing the variability in the process and increasing the product quality. The paper aims to find the low- dimensional information-rich space for soft sensor design using Partial Least Squares-based Multivariate Statistical Process Control technique for the controlling variables in a Wastewater Treatment Plant. The input considered here is the stored data of the actual process variables obtained from the plant. This is carried out using the 14 days data for the three weather conditions, dry rain and storm available from the benchmark model. The performance of the applied method is verified using scatter plot and R-squared.