CNN-LSTM and CNN-GRU based estimation of NOx conversion efficiency in diesel engine exhaust system

Abstract

Reducing nitrogen oxide (NOx) emissions from diesel engines is essential to address environmental pollution and comply with increasingly stringent emission regulations. The significance of this study lies in advancing datadriven solutions for cleaner and more sustainable engine technologies. The aim of this work is to develop and evaluate deep learning models for accurate prediction of NOx conversion efficiency in catalyst-based exhaust after-treatment systems. Four architectures-LSTM, GRU, CNN-LSTM, and CNN-GRU-were implemented using engine load, temperature, and neodymium ratio as input variables, with NOx conversion rate as the target output. Bayesian optimization over 50 iterations was applied to tune hyperparameters, and 5-fold cross-validation was used to ensure robustness. Performance analysis based on MAE, RMSE, and R2 metrics revealed that the hybrid models outperformed individual recurrent networks, with CNN-LSTM achieving the lowest RMSE (0.0727) and highest R2 (0.996). These findings demonstrate the potential of hybrid deep learning approaches for emission control modeling and suggest their applicability in online monitoring systems for efficient NOx reduction without additional sensors.