Machine-Learning-Based Carbon Dioxide Concentration Prediction for Hybrid Vehicles.

Area of Science:

• Environmental Science
• Automotive Engineering
• Computer Science (Machine Learning)

Background:

• Predicting carbon dioxide (CO2) emission concentrations in hybrid vehicles (HVs) is challenging due to dynamic power-train changes.
• Existing machine learning (ML) models have limitations in accuracy, speed, and size for HV CO2 emission prediction.

Purpose of the Study:

• To evaluate traditional and advanced ML models for predicting CO2 emissions in HVs.
• To develop and validate a novel LSTM-based model (UWS-LSTM) for improved CO2 emission prediction.

Main Methods:

• Collected a dataset with over 20 parameters for HV CO2 emissions.
• Conducted extensive input feature optimization to identify key predictive parameters.
• Developed and tested a new long short-term memory (LSTM) model, UWS-LSTM, comparing it against traditional ML and artificial neural network (ANN) models.

Main Results:

• The UWS-LSTM model achieved a prediction accuracy of 97.5%, outperforming existing ML and ANN models.
• The CO2 concentration predictor, implemented on a low-powered IoT device in a commercial HV, demonstrated rapid predictions with an average latency of 21.64 ms.
• The proposed algorithm is computationally efficient, fast, and accurate.

Conclusions:

• The UWS-LSTM model offers a significant advancement in accurately and efficiently predicting CO2 emissions in hybrid vehicles.
• The model's performance on embedded IoT devices highlights its practical applicability in smart vehicle systems.
• This research contributes valuable insights for developing more sustainable and intelligent transportation solutions.