Improvement on the linear and nonlinear auto-regressive model for predicting the NOx emission of diesel engine

Given the increasingly stringent emission regulations, an accurate model of emission prediction is required for the aftertreatment systems of diesel engines. For example, the selective catalytic reduction system can realize higher accuracy emission control if the mass of nitrogen oxides (NOx) is known. Given its simplicity, convenience, and effectiveness, the method of data-driven modeling has been widely researched and considered a primary method to estimate the NOx mass before it reaches the aftertreatment device of a diesel engine. To fully use the known engine operating data and therefore improve the prediction accuracy, this study proposes and develops a general linear and nonlinear auto-regressive model with exogenous inputs (GNARX) for NOx prediction. A recursive least square algorithm with forgetting factor is given to estimate the model parameters, and a new simulated annealing based pruning algorithm is developed to identify the model structure. The proposed methods are first used to model the simulation and engineering data to validate their effectiveness and superiority in comparison to the conventional methods. Based on gray relational analysis, the main factors that influence NOx formation, such as the net engine torque, turbo speed, and accelerator pedal position, are then determined as the inputs for modeling the NOx emission of the diesel engine. The results show that the modeling and prediction accuracy of the GNARX model are higher than those of other models, which indicates that the GNARX model is feasible to predict NOx emission.