Volume 31: Clean Energy Technologies towards Carbon Neutrality

Application of Machine Learning Methods to Predict NOx Emissions for Next Generation Zero-Carbon Ammonia/Hydrogen Fueled Engines Ruomiao Yang, Xiao Zheng, Zhentao Liu, Yu Zhang, Jiahong Fu

Abstract

After industrialization and informatization, the world economy is moving toward “decarbonization”. As two carbon-free fuels, hydrogen and ammonia are attracting increasingly widespread interest around the world. And ammonia and hydrogen are promising and practical alternative energy sources for internal combustion (IC) engines, which may be used to power the next generation of engines due to their zero-carbon footprint. However, the combustion of ammonia produces large amounts of nitrogen oxides (NOx), which pollute the environment. Therefore, it is very significant to control nitrogen oxides (NOx) emissions from spark-ignition (SI) engines. Machine learning (ML) approaches are an alternative analytical tool to three-dimensional (3D) simulations, in-depth experiments and empirical phenomenon models that can accelerate the development of IC engines. The objective of this study was to assess the applicability of ML models in predicting NOx emissions. A calibrated spark-ignition engine fueled with gasoline operating under different conditions was used to provide sufficient data for model training, validation, and testing. The results indicated that the artificial neural network (ANN) and support vector regression (SVR) have good prediction performance and high accuracy, and the prediction accuracy of the RF model is acceptable. In general, ANN and SVR have comparable performance and both models are recommended to predict NOx emissions from ammonia/hydrogen fueled engines. And the prediction performance of the RF model will be less accurate compared to the other two ML models.

Keywords Machine learning, NOx, ammonia/hydrogen, artificial neural network, support vector regression, random forest