Statistical Analysis of City Bus Driving Cycle Features for the Purpose of Multidimensional Driving Cycle Synthesis 2020-01-1288
Driving cycles are typically defined as time profiles of vehicle velocity, and as such they reflect basic driving characteristics. They have a wide application from the perspective of both conventional and electric road vehicles, ranging from prediction of fuel/energy consumption (e.g. for certification purposes), estimation of greenhouse gas and pollutant emissions to selection of optimal vehicle powertrain configuration and design of its control strategy. In the case of electric vehicles, the driving cycles are also applied to determine effective vehicle range, battery life period, and charging management strategy. Nowadays, in most applications artificial certification driving cycles are used. As they do not represent realistic driving conditions, their application results in generally unreliable estimates and analyses. Therefore, recent research efforts have been directed towards development of statistically representative synthetic driving cycles derived from recorded GPS driving data. The state-of-the-art synthesis approach is based on Markov chains, typically including vehicle velocity and acceleration as Markov chain states. However, apart from the vehicle velocity and acceleration, a road slope and vehicle mass are also shown to significantly impact characteristics of driving cycle, and thus should be included into synthesis. The vehicle mass is particularly important for delivery and urban transport vehicles (e.g. trucks and city buses), where mass can vary significantly depending on delivered cargo in the case of trucks or number of passengers in the case of buses.
This paper deals with preprocessing and statistical analysis of recorded GPS driving data for the purpose of multidimensional driving cycle synthesis. The data have been collected for a city bus fleet operating in the city of Dubrovnik over a wide period of time. The vehicle acceleration, road slope, and vehicle mass, which are important from the standpoint of driving cycles synthesis along with vehicle velocity (i.e. input variables), were obtained by preprocessing recorded raw data. The road slope profiles were obtained by using regression technique based on Gaussian processes, while the vehicle mass profiles are estimated from standard CAN and GPS data and verified against the data obtained by counting passengers on selected bus routes. An important step in driving cycle synthesis is calculation of a joint conditional probability distribution over considered input variables. In order to reveal appropriate and computationally efficient way of calculating this distribution, comprehensive statistical analysis of individual input variables and their mutual correlations are conducted. Additionally, in order to account for seasonal and daily variation in driving characteristics, the recorded driving data is clustered by using k-means algorithm into appropriate number of distinctive groups. Each group is meant to be represented by a separate synthetic driving cycle. Finally, a procedure of synthesis of multidimensional driving cycle will be demonstrated and analyzed.
Jakov Topić, Branimir Skugor, Josko Deur