Prediction of NOx Emission from Coal Fired Power Plant Based on Real-Time Model Updates and Output Bias Update

Abstract

In order to deal with the nonlinear varying behavior of NOx emissions for long term predictions, a real-time recursively updating model is indispensable. In this paper, new recursively updating models are proposed to predict NOx emissions. The proposed real-time models are equipped with an initial LSSVM model and subsequent updating methods to adapt the models with recent changes to process data. The updating methods include solo Least Squares Support Vector Machines (LSSVM) update, solo output bias update, and the combination of these two termed as the LSSVM-Scheme. These models are applied to NOx emission process data from a coal combustion power plant in Korea. Prediction results obtained from the proposed real-time LSSVM models are compared with their counterpart real-time PLS models, which reveal that real-time LSSVM models outperform their counterpart real-time PLS models. Among other models developed in this work, LSSVM-Scheme and solo output bias update based on LSSVM predicts NOx emissions robustly for a long passage of time with the highest accuracy.