Nitrogen Oxides Concentration Estimation of Diesel Engines Based on a Sparse Nonstationary Trigonometric Gaussian Process Regression With Maximizing the Composite Likelihood

Gaussian process regression (GPR) has been a popular Bayesian method for nonlinear fitting. It has the advantage of predictive capability, uncertainty measurement, and interpretable structure. However, the original GPR has a heavy complexity, which limits its effectiveness on Big Data problems. Though plenty of sparse GPR methods were proposed to deal with it, they usually result in reducing the prediction accuracy. In this article, a novel sparse GPR with a newly defined objective function is proposed to obtain the hyperparameters in a different manner compared to traditional maximizing-likelihood. Experimental results on a real diesel engine dataset and several public datasets verify that the proposed method can have a better performance on the prediction.