Fuel consumption and CO₂ emissions prediction in road transport using a hybrid deep learning approach

Abstract

In recent times, due to the substantial increase in population and the number of vehicles, road transportation has emerged as a crucial factor contributing to energy consumption and the release of CO2 emissions. In road transportation, the analysis of driver behavior is one of the most significant approaches to evaluating vehicle fuel consumption (FC) and CO2 emissions. Driver behavior analysis provides valuable insights into the influence of driving habits and characteristics on FC and CO2 emissions, including factors such as driving style, average vehicle speed, and acceleration pedal usage. This analysis provides a comprehensive understanding of how these factors impact real-world fuel efficiency. For this purpose, this study utilizes two non-linear methods, an artificial neural network (ANN) and a hybrid deep learning method called long short-term memory with multi-verse optimizer (LSTM-MVO), to predict FC and CO2 emissions. In this study, 370 data points were recorded, and relevant driver behaviors’ parameters were measured from mobile technology and the OBD interface for FC and CO2 emissions, respectively. The input data for modeling and determining an optimized function for predicting the amount of fuel consumed and CO2 emissions include Fuel_flow_rate/hour (FFRH), Engine_RPM (ERMP), Speed (S), Acceleration (A), and Grade (G). Also, the amount of FC and CO2 emissions are also considered outputs data. The results clearly showed that the LSTM-MVO approach can provide higher performance capacity in predicting fuel consumption and CO2 emissions compared to ANN. Finally, the results of this study emphasize its potential as a promising approach to address specific issues related to driver behavior and environmental pollution.