Evaluating Artificial Neural Networks and Emerging Machine Learning Methods for Engine Performance and Emissions Prediction: A Review

These due to the shift towards the cleaner energy generation and newer and sophisticated engine technologies the prediction of the performance and emission of engine is of vital importance for the designer for the purpose of ANR as well as the emission control standards. This review also assesses the performance of ANNs and the new machine learning techniques in this area. ANNs especially the single layer and multi-layer networks have shown good levels of predictions of engine performance and emission for ethanol-gasoline, bioethanol and other fuel blends. For example, the models based on Levenberg-Marquardt optimization and the Bayesian regularization have revealed essential correlation coefficients and minimized mean squared errors. Based on comparative analyses, the further improvement of the predictive ability and speed of ANNs is stressed, including applications of ANNs integrated with other approaches like Response Surface Methodology (RSM), and other developing approaches like Extreme Learning Machines (ELM) and Support Vector Regression (SVR). However, challenges are concluded to remains including data quality, model complexities and computational resources. This review reveals that, for tackling these issues and enhancing the progress in developing efficient and sustainable engine solutions, detailed databases and refined mathematical modeling approaches are required. The directions for future research are to obtain better quality data, enhance computational time, and determine more fuel blends and engine settings for the model.