Expertise-guided NOx emission modeling of hybrid vehicle engines via peak-valley-enhanced Gaussian process regression

In order to improve the characterization accuracy of NOx emission of hybrid vehicle engines, this paper proposes an expertise-guided NOx emissions modeling method of peak–valley-enhanced Gaussian process regression (PV-GPR) to precisely capture its mapping features. A K-nearest neighbors model is applied first to classify data based on engine operating conditions defined by the experts. Customized Gaussian process regression models are then developed for NOx emission under each condition. Each GPR model features a customized kernel function with identified peak and valley positions. All data was collected from an experimental test bench with a BYD gasoline engine for hybrid vehicles. The results show that using the proposed PV-GPR method achieves a lower RMSE (0.49), significantly outperforming the feedforward neural network (0.84) and cascade neural network (1.01). Gaussian kernel functions applied to NOx modeling in hybrid vehicle engines are designed, further extending the method’s applicability. • An expertise-guided NOx emissions modeling method is proposed. • This method riches peak–valley natures of Gaussian process regression. • Test bench experiments were implemented for method validation. • Gaussian kernel functions applied to NOx emission modeling are designed.