A robust, data-driven methodology for real-world driving cycle development

摘要

This paper develops a robust, data-driven Markov Chain method to capture real-world behaviour in a driving cycle without deconstructing the raw velocity-time sequence. The accuracy of the driving cycles developed using this method was assessed on nine metrics as a function of the number of velocity states, driving cycle length and number of Markov repetitions. The road grade was introduced using vehicle specific power and a velocity penalty. The method was demonstrated on a corpus of 1180 km from a trial of electric scooters. The accuracies of the candidate driving cycles depended most strongly on the number of Markov repetitions. The best driving cycle used 135 velocity modes, was 500 s and captured the corpus behaviour to within 5% after 1,000,000 Markov repetitions. In general, the best driving cycle reproduced the corpus behaviour better when road grade was included.