International Symposium of Transport Simulation & International Workshop on Traffic Data Collection and its Standardization 2018

1 / 2017-12-01 14:36:42

Recurrent Neural Network Based Driving Cycle Development for Light Duty Vehicles in Beijing

Driving Cycle,Light Duty Vehicle,Recurrent Neural Network,Mean Tractive Force

This paper presents a data driven methodology for developing driving cycles with deep recurrent neural network (DRNN) architecture. In contrast to the approaches that features sub-cycle selection from the original data set through Markov process and data clustering, our method models the time-variant discrete distributions of velocities and generates driving cycles step by step with the model instead . Such data driven approach excludes the necessity for extracting features with domain knowledge during the modeling stage. In the end of the paper our method is validated with comparisons between synthesized driving cycle and the original dataset based on 14 metrics. As the final result, the driving cycle obtained features relatively high power demand compared with the original data set.