Driving behavior recognition and fuel economy evaluation for heavy-duty vehicles

Driving Heavy-Duty Vehicles (HDVs) is inherently energy-intensive, making it significant for energy conservation and emission reduction in transportation. While prior research has acknowledged the influence of driving behavior on fuel consumption and analyzed it using statistical approaches, limited attention has been given to refining drivers' behavior through tailored and differentiated strategies to further promote eco-driving. Considering real-world and real-time monitoring scenarios, this study proposes an offline training and online service framework to provide specific and quantifiable strategies for reducing fuel consumption in HDV trips. During the offline phase, the Toeplitz Inverse Covariance Clustering (TICC) algorithm is employed to segment and recognize driving behaviors using historical HDV data. Building upon this behavior recognition, we integrate multiple sources of factors to develop a model linking them to fuel consumption and conduct a qualitative analysis of their contributions. In the online service phase, the trained TICC model maps and identifies real-time driving behaviors during trips. Meanwhile, a multi-objective counterfactual explanation model generates feedback strategies that consider personalized requirements for fuel consumption reduction.