Enhancing waste cooking oil biodiesel yield and characteristics through machine learning, response surface methodology, and genetic algorithms for optimal utilization in CI engines

Abstract

ABSTRACT The current study seeks to predict and optimize the biodiesel production yield and physicochemical properties of waste cooking oil. Using Box-Behnken design (BBD), L46 orthogonal arrays were developed for experimentation at five factors and three levels. Furthermore, three different boosting algorithms (AdaBoost, ExtraTrees, and Gradient-Boosting regression) were used to develop an ML-based prognostic model using experimental data. Based on the coefficient of determination (R2) (0.997 for biodiesel yield, 0.999 for kinematic viscosity, 0.996 for calorific value, and 0.999 for flash point), the Gradient-Boosting model has the most accurate predictions, followed by ExtraTrees and AdaBoost. Through the implementation of GA, the most optimal conditions for biodiesel production were obtained, including a molar ratio of 7.24:1, catalyst concentrations of 1.49 wt.%, reaction temperatures of 65°C, reaction times of 59.95 minutes, and stirring speeds of 733.32 rpm. Experimental validation of these optimized conditions is closely aligned with the predicted values. Furthermore, the study evaluates engine performance and emission parameters to assess the impact of different biodiesel/diesel blends (B10, B20, and B30) compared to pure diesel. The utilization of these biodiesel blends demonstrates satisfactory performance, effectively reducing carbon monoxide (CO) and hydrocarbon (HC) emissions. However, nitrogen oxide (NOx) emissions increase compared to pure diesel.