Support Vector Machine-based Design of Multi-model Inferential Sensors

Abstract

There is an increasing need to monitor industrial key variables by inferential (soft) sensors. This contribution deals with the challenge of increasing the accuracy of inferential sensors yet maintaining the simple (linear) structure. In order to fulfill these opposing requirements, we design a linear multi-model inferential sensor (MIS) that switches between two models. We enhance the design of a sensor by continuous switching and optimized data labeling. The case study deals with a nonlinear model of pressure-compensated temperature often used in distillation columns monitoring and control. The results show a significant accuracy improvement of MIS over a single-model sensor. The studied MIS design approaches present a great potential for practical use.