Scholarlink

<p>The environment is facing numerous challenges, and one of the major challenges is excessive reliance on fossil fuels. Fuel Efficiency can be most accurately predicted from the sensor inputs from on-board computers. An attempt is made to predict the efficiency of an internal combustion in a vehicle via taking data from the on-board Electronic Control Unit (ECU). A hybrid framework has been proposed which utilizes a data pipeline to ensure which accepts data in real-time from the ECU via On Board Diagnostics (OBD2), moreover the model performs a prediction and return efficiency figures. This research proposes a real-time ensemble-based machine learning technique to generate a system for fuel efficiency measurement. Accuracy is preserved on addition of new data crucial for long-term stability of the model by implementing drift controlling measures. Open-source datasets coupled are used for training base model and raw data from dynamometer and OBD2 port of vehicle for simulating real conditions. This demonstrates capability of data processing pipeline which continually improves the model capabilities. The model achieves a low Mean Absolute Error (MAE) of 0.048, a Root Mean Square Error (RMSE) of 0.0693, which demonstrate its high accuracy and a high R2-Score of 0.9729 indicating strong precision and goodness-of-fit. Being CPU focused allows its deployment in low-power embedded ECU hardware of any modern vehicle.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Realtime Fuel Efficiency Prediction Using Stacking Ensemble Learning

  • 期刊论文

  • 发表日期：

    2026年1月28日

查看全文
  • Sanjit Kumar Dash,
  • Arbin Mahapatra,
  • Anwesh Choudhury,
  • Muktikanta Sahu,
  • Soubhagya Ranjan Mallick

摘要

The environment is facing numerous challenges, and one of the major challenges is excessive reliance on fossil fuels. Fuel Efficiency can be most accurately predicted from the sensor inputs from on-board computers. An attempt is made to predict the efficiency of an internal combustion in a vehicle via taking data from the on-board Electronic Control Unit (ECU). A hybrid framework has been proposed which utilizes a data pipeline to ensure which accepts data in real-time from the ECU via On Board Diagnostics (OBD2), moreover the model performs a prediction and return efficiency figures. This research proposes a real-time ensemble-based machine learning technique to generate a system for fuel efficiency measurement. Accuracy is preserved on addition of new data crucial for long-term stability of the model by implementing drift controlling measures. Open-source datasets coupled are used for training base model and raw data from dynamometer and OBD2 port of vehicle for simulating real conditions. This demonstrates capability of data processing pipeline which continually improves the model capabilities. The model achieves a low Mean Absolute Error (MAE) of 0.048, a Root Mean Square Error (RMSE) of 0.0693, which demonstrate its high accuracy and a high R2-Score of 0.9729 indicating strong precision and goodness-of-fit. Being CPU focused allows its deployment in low-power embedded ECU hardware of any modern vehicle.