Extended Fast Relevance Vector Regression based Pollutant Concentrations Prediction for Biomass Cogeneration Systems

Abstract

Accurate and reliable prediction of pollutant emission concentrations from biomass cogeneration systems is critical to improving energy efficiency and reducing environmental pollution. The Relevance Vector Regression (RVR) algorithm, with its strong ability to represent stochastic uncertainty, has become an effective method for pollutant concentration prediction in biomass cogeneration systems. However, the classical RVR algorithm is mainly used for univariate prediction and has good prediction results only for small sample data. In order to address the problems of biomass cogeneration systems with far more than one pollutant and large data sets, a prediction model based on an improved Fast Relevance Vector Regression (FRVR) algorithm is proposed in this paper. Specifically, the model is divided into two parts with different methods: a K-means method to partition the dataset and a prediction based on an improved FRVR algorithm. First, the K-means method is used to divide the large data set into smaller data sets so that the prediction model can better extract all the useful information from the original data. Second, the FRVR algorithm is extended to multivariate output to achieve simultaneous prediction of multiple pollutant concentration. Finally, the experimental results verified that the proposed algorithm has great performance in the pollutant concentrations prediction of biomass cogeneration systems.