Development of machine-learning models to determine the power of natural gas/hydrogen spark ignition engines

Abstract

This study develops a set of machine learning models for predicting the power output of a spark ignition engine operating on natural gas enriched with hydrogen. The input variables considered in the modeling framework include the hydrogen blending ratio (%), adiabatic index, in-cylinder temperature (K), and pressure (bar) during the compression stroke. The dataset used for model training and validation was obtained from experimental measurements conducted on a spark-ignition engine using a standardized test bench. The machine learning techniques employed in this work encompass Linear Regression (LR), Linear Regression with Interaction Terms (LR-INTER), Fine Tree (FTREE), Ensemble Learning (ENS), Exponential Gaussian Process Regression (EGPR), Cubic Support Vector Machine (CSVM), Fuzzy Logic (FL), and Artificial Neural Networks (ANN). Model performance was assessed using the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). Among all evaluated techniques, the LR-INTER model delivered the highest predictive accuracy, achieving R² = 0.99999, RMSE = 3.35 × 10⁻⁵, and MAE = 2.94 × 10⁻⁵ on the testing dataset. In addition, a Graphical User Interface (GUI) was developed to integrate the trained models and facilitate practical use. This interface provides researchers and engineers with an effective tool for instructional purposes and for forecasting the power characteristics of hydrogen-assisted natural gas spark ignition engines.