Application of supervised machine learning and Taylor diagrams for prognostic analysis of performance and emission characteristics of biogas-powered dual-fuel diesel engine

In the ongoing search for an alternative fuel for diesel engines, biogas is an attractive option. Biogas can be used in dual-fuel mode with diesel as pilot fuel. This work investigates the modeling of injecting strategies for a waste-derived biogas-powered dual-fuel engine. Engine performance and emissions were projected using supervised machine learning methods including random forest, lasso regression, and support vector machines (SVM). Mean Squared Error (MSE), R-squared (R²), and Mean Absolute Percentage Error (MAPE) were among the criteria used in evaluations of the models. Random Forest has shown better performance for Brake Thermal Efficiency (BTE) with a test R² of 0.9938 and a low test MAPE of 3.0741%. Random Forest once more exceeded other models with a test R² of 0.9715 and a test MAPE of 4.2242% in estimating Brake Specific Energy Consumption (BSEC). With a test R² of 0.9821 and a test MAPE of 2.5801% Random Forest emerged as the most accurate model according to carbon dioxide (CO₂) emission modeling. Analogous results for the carbon monoxide (CO) prediction model based on Random Forest obtained a test R² of 0.8339 with a test MAPE of 3.6099%. Random Forest outperformed Linear Regression with a test R² of 0.9756% and a test MAPE of 7.2056% in the case of nitrogen oxide (NOx) emissions. Random Forest showed the most constant performance overall criteria. This paper emphasizes how well machine learning models especially Random Forest can prognosticate the performance of biogas dual-fuel engines.