Journal
APPLIED ENERGY
Volume 298, Issue -, Pages -Publisher
ELSEVIER SCI LTD
DOI: 10.1016/j.apenergy.2021.117204
Keywords
Electric bus energy consumption; Stochastic speed profile generation; Machine learning; Knearest neighbour searching; Big data
Categories
Funding
- National 135 Key R&D Program Projects [2018YFB1600600]
- Basic Research Program of Shenzhen Science and Technology Innovation Committee [JCYJ20180307123910003]
- Scientific Research Funds of Tsinghua Shenzhen International Graduate School
Ask authors/readers for more resources
A novel two-step EBEC prediction methodology, incorporating a stochastic RF model and kNN RF model, is proposed to improve the prediction accuracy of electric bus energy consumption. The models show great performance in real-world data collected from over 163,800 journeys in Shenzhen, China, with the stochastic RF model achieving a significant increase in prediction accuracy and the kNN RF model reducing computation time while maintaining accuracy.
With increasing adoption of electric buses, an accurate and real-time electric bus energy consumption (EBEC) prediction can significantly contribute to the battery capacity determination and dynamic scheduling management. In this study, a novel two-step EBEC prediction methodology including a stochastic random forest (RF) model, and a cooperative k nearest neighbour (kNN) RF model has been proposed to improve the prediction accuracy and achieve faster response. The stochastic RF model is developed to enhance prediction ability for the EBEC of the terminus-to-terminus journey by integrating stochastic speed profile generation for kinematic features extraction with RF algorithm for considering external factors as input parameters. Based on this, the kNN RF model is built to substitute the stochastic generation process in order to boost forecasting efficiency after a certain amount of trip accummulation. The models have been performed on real-world data collected from over 163,800 journeys over five consecutive months in Shenzhen, China. The results show that the proposed stochastic RF model has a great prediction performance with all the selected indicators, and the accuracy in MAE is raised by 9.470-51.144% compared with the existing models. In addition, the kNN RF model can save computation time by up to 97.142% with accuracy reduction of 1.795% in MAE compared with stochastic RF model. This means that the kNN searching model can be effectively used to reduce great computational cost with accumulation of trip scenarios by sacrificing very little accuracy. The developed models can be applied to the city-scale electric bus operation system, to benefit real-time scheduling and management.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available