Machine Learning Techniques Improve Predictive Models for Emissions in Gasoline-Alcohol Blends

Strategic Insights -

Scientists at Sichuan University have employed machine learning techniques to develop more accurate predictive models for emissions in gasoline-alcohol blends. By using Gradient Boosting, Random Forest, Bootstrap Aggregating, and Extreme Gradient Boosting (XGB) algorithms, the researchers aimed to improve the accuracy of predictions for carbon monoxide, unburned hydrocarbons, and nitrogen oxides emissions. The results of the study showed that the XGB model provided the best accuracy, with absolute average relative deviation percentages of 2.997%, 2.184%, and 8.370% in predicting carbon monoxide, unburned hydrocarbons, and nitrogen oxides emissions, respectively.

Key Takeaways:

  • The researchers investigated the feasibility of using machine learning techniques to predict emissions in gasoline-alcohol blends, which are harmful to human health and the environment.
  • The study demonstrated that the XGB model provided the best accuracy in predicting emissions, with significant reductions in absolute average relative deviation percentages compared to previous studies.
  • The key contribution of this study lies in the superior generalization and prediction performance of the XGB framework compared to previously developed machine learning models in the literature.
  • The research teams' model showed strong correlations between emissions and fuel properties, engine operating parameters, and blend composition.
  • The study employed a grid search method to determine the hyperparameters of the machine learning algorithms, providing a comprehensive methodology for designing the topology of predictive models.

Statistics:

  • The XGB model provided the best accuracy in predicting emissions, with absolute average relative deviation percentages of 2.997%, 2.184%, and 8.370% for carbon monoxide, unburned hydrocarbons, and nitrogen oxides emissions, respectively.
  • The study employed a grid search method to determine the hyperparameters of the machine learning algorithms, demonstrating a multifaceted approach to predictive modeling.
  • The study involved researchers from the Sichuan University, Jinjiang College, and included additional authors Mingyang Wang, Huiyi Cui, and Aotian Wang.

Sources:

  • VerticalNews (2025)
  • Applying and Validating Ai-powered Frameworks for Estimating Emissions of Alcohol/gasoline Combustion In Internal Engines. Journal of Environmental Chemical Engineering, 2025;13(5).
  • Sichuan University (2025).