Explainable Prognostics-optimization of Hydrogen Carrier Biogas Engines in an Integrated Energy System using a Hybrid Game-theoretic Approach with XGBoost and Statistical Methods

Biogas is a renewable fuel source that helps the circular economy by turning organic waste into energy. This study tackles existing research gaps by exploring the use of biogas as a hydrogen carrier in dual-fuel engine systems. It additionally employs explainable machine learning techniques for predictive modelling and interpretive analysis. The dual-fuel engine was powered with biogas as main fuel while biodiesel-diesel blend was used as pilot fuel. The engine was tested at different Compression Ratios (CR) and Brake Powers (BP). The generated data from testing was used to develop the mathematical models and parametric optimization of engine performance and emissions using Response Surface Methodology (RSM). Desirability-based optimization identified optimal results: a Peak Cylinder Pressure (Pmax) of 54.97 bar and a brake thermal efficiency (BTE) of 24.35 %, achieved at a CR of 18.3 and a BP of 3.3 kW. The predictive machine learning approach, Extreme Gradient Boosting (XGBoost), was employed to develop predictive models. XGBoost precisely forecasted engine performance and emissions, with Coefficient of Determination (R2 ) values (up to 0.9960) and minimal Mean Absolute Percentage Error (MAPE) values (1.47–4.89 %) for all parameters. SHapley Additive exPlanations (SHAP) based analysis identified BP as the predominant feature with a normalized importance score reaching up to 0.9, surpassing that of CR. These findings underscore the potential of biogas as a viable, sustainable fuel and highlight the role of explainable prediction–optimization frameworks can play in achieving optimal engine performance and emission control.