A Knowledge-and-data-driven Soft Sensor for Predicting Air Preheater Rotor Deformation

Abstract

Soft sensors have been widely used in industrial processes for the past two decades. These sensors use easy-to-measure process variables to predict hard-to-measure ones. This paper proposes a novel knowledge-and-data-driven modeling (KDDM) approach for soft sensors. One submodel is derived from all available domain knowledge, including all known relationships from physically based or mechanistic models; the other is constructed solely from data without using any domain knowledge. The two submodels were integrated with a mass balance model. In this work, a mathematical model based on technical expertise was adopted as the knowledge-driven submodel, and the support vector machine regression (SVR) model was adopted as the data-driven submodel. To improve the model’s performance, the genetic algorithm (GA) is used to obtain the optimal parameters of the SVR. For the purpose of evaluating the proposed method, a soft sensor model that estimates the rotor deformation of an air preheater in a thermal power plant boiler is studied. The experimental results demonstrate that, in comparison with the common knowledge-driven model (KDM) and data-driven model (DDM), the proposed KDDM approach is able to provide higher predictive accuracy for the rotor deformation of the air preheater. The case study confirms that the KDDM approach inherits advantages from both the KDM and DDM approaches.