Soft Sensor for Prediction of High-temperature Corrosion of Boiler Tubes: Formulation and Validation of the Machine Learning Model

Abstract

A soft sensor is developed to predict high-temperature corrosion rates of boiler tubes from plant operation data. A machine learning model of the soft sensor is formulated in combination of multiple linear regression analysis and principle component analysis. Model parameters of the prediction formula with plant operation data as variables are optimized by measured data of corrosion rate using an electrochemical probe, and the generalized performance is evaluated with cross-validation.