Artificial neural networks vs. gene expression programming for predicting emission & engine efficiency of SI operated on blends of gasoline-methanol-hydrogen fuel

Abstract

While retaining environmental friendliness, robust modelling and enhancing spark ignition engine efficacy can be done using improved innovative fuel and unconventional robust hybrid tools. This study is the first to employ Al techniques such as artificial neural networks (ANN) and gene expression programming (GEP) to predict the performance and emissions of a gasohol/hydrogen-powered SI engine. The ANN was adopted to correlate the engine variables viz. engine speed and gasohol/hydrogen mix vs. responses namely brake thermal efficiency (BTE), brake specific energy consumption (BSEC), carbon monoxide (CO), hydrocarbon (HC), oxides of nitrogen (NOx) and carbon dioxide (CO2). GEP model was further employed to predict BTE, BSEC, CO, HC, NOx and CO2. To examine the prediction efficacy of both AI techniques, a set of advanced statistical approaches was used. A set of advanced statistical techniques were employed to test the prediction efficiency of both AI techniques. It was revealed that ANN outperformed the GEP since the values for R in the case of ANN were 0.9864–0.9998 whereas the values for R in the case of GEP were 0.9864–0.9994. Similarly, in the instance of R2, ANN outperformed GEP. Furthermore, Kling-Gupta efficiency was greater in the case of ANN (0.9684–0.9999) than in GEP (0.8912–0.9991). Both AI approaches, however, displayed great prognostic effectiveness in forecasting engine performance and emissions.

Keywords

• ANN
• Gene expression programming
• Modelling
• Gasohol
• Hydrogen
• Spark ignition engine