Spectral clustering based approach for evaluating the effect of driving behavior on fuel economy

Measurement or evaluation of driving behavior is vital to the development of advanced driver assistance systems (ADAS). Accurate assessment of a driver's behavior allows the ADAS to make appropriate recommendations and function in harmony with the driver by adapting to the driver's personal driving style. Traditional driving behavior analysis methods focus on driving process modeling, which is complex and rarely be generalized for some special application like fuel consumption evaluation. In particular, the complexity of the modeling process, in combination with dynamic driving conditions, makes the quantitative analysis of the relationship between driving behavior and fuel consumption difficult to evaluate. Therefore, in this paper we propose a data mining-based method which can be used to measure the driving behavior's effect on fuel consumption by using real-world driving data. First, we introduce a spectral clustering algorithm which will be used with driving data to group drivers according to their driving styles. Then, we design a driving behavior evaluation experiment with fixed traffic conditions and a wireless data collection platform to collect the data. A data processing mechanism is also proposed to obtain the proper experimental driving data suitable to be the spectral clustering algorithm. Finally, we use actual fuel consumption as the ground truth to verify the accuracy of our method. The results show that the proposed method can measure the driving behavior's effect on the vehicle's fuel consumption well.