Premium
Artificial intelligence‐based model prediction of biodiesel‐fueled engine performance and emission characteristics: A comparative evaluation of gene expression programming and artificial neural network
Author(s) -
Sharma Prabhakar
Publication year - 2021
Publication title -
heat transfer
Language(s) - English
Resource type - Journals
eISSN - 2688-4542
pISSN - 2688-4534
DOI - 10.1002/htj.22138
Subject(s) - gene expression programming , artificial neural network , correlation coefficient , diesel fuel , diesel engine , biodiesel , coefficient of determination , automotive engineering , environmental science , computer science , engineering , machine learning , chemistry , biochemistry , catalysis
The prognostic capability of gene expression programming (GEP) and artificial neural network (ANN) are compared to estimate the engine performance and emission characteristics. A stationary diesel engine was powered with linseed oil biodiesel‐mineral diesel blends. A total of 60 lab‐based test‐run were conducted by varying the engine input operating conditions, namely fuel injection parameters, diesel/biodiesel blending ratio, and engine load. The engine output data, namely brake thermal efficiency and brake‐specific fuel consumption, were calculated, while emission data for oxides of nitrogen, carbon monoxide, and unburnt hydrocarbon, were recorded. The experimental data were used for predictive model development using artificial intelligence‐based GEP and ANN techniques. The developed models were tested on statistical outcomes, such as the absolute fraction of variance (0.9698–0.997 for GEP and 0.9949–0.9998 for ANN), correlation coefficient (0.9848–0.998 for GEP and 0.9974–0.9998 for ANN), establishing these two models as an efficient machine identical tool. Also, Nash–Sutcliffe efficiency (0.937–0.9999 for GEP and 0.995–0.999 for ANN) and Kling–Gupta efficiency (0.834–0.9999 for GEP and 0.989–0.999 for ANN) elevate the prediction quality of developed models. The result showed that the ANN model was slightly more accurate than the GEP‐based model for the same parametric range.