Analysis of Markov Chain-based Methods for Synthesis of Driving Cycles of Different Dimensionality

Driving cycles reflect driver behavior and local traffic characteristics, and they are often described in terms of vehicle velocity time profile and widely used for different certification purposes. In order to obtain reliable fuel consumption estimates, realistic driving cycles including the road slope profile should be used, especially for electric vehicles whose energy consumption may significantly vary depending on road and traffic conditions. As a continuation of previous research related to statistical analysis of a rich set of city bus driving cycles, this paper focuses on multidimensional driving cycle synthesis using the Markov chain method. Five Markov chain models are derived with respect to different selections of states and correspondingly different dimensionalities of transition probability matrix, and they are systematically compared against certain statistical criteria. In order to reduce memory demand while improving computational efficiency, a sparse implementation of transition probability matrix based on dictionaries of keys is employed. Finally, the analyzed synthesis methods are used to generate a small set of synthetic driving cycles that represent the recorded driving cycles in statistically reliable sense.