Comparison of Different Variable Combinations for Electric Vehicle Power Prediction Using Kernel Adaptive Filter. Issue 20 (2021)

• Record Type: Journal Article
• Authors: Shen, Heran; Wang, Zejiang; Yang, Kuo; Lamantia, Maxavier; Chen, Pingen; Wang, Junmin
• Is Part Of: IFAC-PapersOnLine. Volume 54:Issue 20(2021)
• Page Start: 858
• Page End: 863
• Publication Date: 2021
• Subjects: Kernel adaptive filters -- longitudinal dynamics model -- power prediction -- electric vehicle; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2021.11.279
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: Range anxiety has always been one of the critical issues affecting electric vehicle commercial penetration and customer acceptance. Current electric vehicle range estimation methods generally fall into two categories: model-based methods and data-driven methods. The former requires all vehicle-specific parameters, whereas the latter relies on past energy consumption data. This paper integrates the advantages of the two approaches. It compares three different combinations of variables used as inputs for online machine-learning methods to identify the vehicle's instantaneous power consumption. The vehicle longitudinal dynamic model inspires the proposed variable combinations. Collected signal data are preprocessed by algebraic derivative estimation to filter noise and smooth both the original data and the data's first-order derivative. The performance of the proposed variable combinations is evaluated by six datasets. The results indicate that one of the newly proposed combinations has better accuracy than all other varieties.