Machine learning models for biogas potential in sustainable aviation: XGBoost, random forest and ridge regression

This study aims to focus on the performance and emissions characteristics of different combinations of biofuel blends in aviation engines using machine learning models. The paper discusses both energy performance and emissions reduction, so it can be clarified in the title and abstract.

The blends tested were B10 (10% microalgae, 90% Jet A fuel), BB10 (10% biodiesel, 10% biogas, 80% Jet A fuel), B30 (30% microalgae, 70% Jet A fuel) and BB30 (30% biodiesel, 10% biogas, 60% Jet A fuel), respectively. All the blends are tested already in the previous study, and the results were trained using the ML models here, and the comparison was made.The machine learning models used were XGBoost, random forest and ridge regression. These models were trained using the actual data of thrust, thrust specific fuel consumption (TSFC), turbine inlet temperature (TIT), nitrogen oxides (NOx) emissions, carbon monoxide (CO) emissions and carbon dioxide (CO₂) emissions. Trained models were evaluated using experimental data, and their performance is assessed based on root mean squared error, mean absolute error and R-squared (R²) metrics.

From the results, it is clear that the random forest model emerges as the most effective in predicting thrust, TIT and CO₂ emissions by reporting low error and high R². On the other hand, the ridge regression model outperforms other models in predicting TSFC, NOx emissions and CO emissions. Considering all results, most models capture the movements of reduced thrust, increased TSFC and slightly higher TIT. Meanwhile, the models have the ability to capture the lower NOx, CO and CO₂ emissions for biodiesel blends compared to Jet A fuel. The study also specifying that the models are used for regression would add clarity since the study focuses on predicting performance and emissions characteristics using continuous numerical outputs.

Based on all the predictions from the trained models, it is clear that the machine learning models can support understanding the performance and emissions characteristics of biodiesel blends and support decision-making processes in fuel selection and engine performance optimization.

The main objective of the study is to provide insights into the potential of biodiesel blends as alternative fuels in aviation using various machine learning models in predicting critical aviation parameters, including thrust, TSFC, TIT and emissions.