Determination of the electric vehicles driving modes in real life conditions by classification methods

Ben-Marzouk, Mohamed; Clerc, Guy; Pélissier, Serge; Sarı, Ali; Venet, Pascal · 2018 · OpenAlex-citations

DOI: 10.1109/icit.2018.8352506

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of characterizing real-world driving modes for electric vehicles (EVs) to better understand battery aging mechanisms. While battery lifetime predictions often rely on standardized cycles like NEDC or WLTC, these simplified profiles fail to capture the complexity and diversity of actual usage conditions, such as temperature variations and state of charge fluctuations. The authors argue that analyzing real-life data is essential for accurate lifespan modeling and cost-effective battery dimensioning. The study utilizes a database from the CROME project, comprising data from ten EVs equipped with data loggers that recorded over 500 variables at a 0.01-second frequency. The dataset includes parameters related to vehicle operation (speed, acceleration), battery status (current, voltage, temperature), and engine metrics. To manage this high-dimensional data, the authors first reduced the variable set from over 500 to 140 by eliminating irrelevant factors, such as engine cooling temperature. They then applied Pearson and Spearman correlation coefficients to identify and remove redundant variables, retaining only one variable per highly correlated set. Shannon’s entropy was used to select the most informative variable within each correlated group, resulting in 18 significant input variables. Although Laplacian score analysis was attempted for further feature selection, it indicated that all 18 variables held significant information, so none were discarded. To facilitate classification, Principal Component Analysis (PCA) was employed for dimension reduction. While statistical criteria suggested four components, the authors selected seven principal components to ensure that the projection quality for all variables exceeded 75%. The driving cycles were then classified using unsupervised k-means clustering, which was chosen over hierarchical ascending classification for its computational efficiency and ability to update class centers. The analysis determined that five distinct classes optimally explained the variance in the data. The results identified five distinct driving modes. Modes 1 and 2 were characterized as urban driving cycles, distinguished by lower average and maximum speeds. Modes 4 and 5 represented highway driving, featuring higher speeds. Within these categories, differences in aggressiveness were noted: Mode 4 exhibited more aggressive behavior than Mode 5, and Mode 2 was more aggressive than Mode 1, as indicated by higher relative positive acceleration and kinetic energy metrics. Conversely, variables such as charge consumed per kilometer and ambient temperature showed little variation across modes, suggesting these factors are not primary differentiators of driving style. The authors conclude that this methodology provides a robust framework for linking specific real-world usage patterns to battery aging rates, offering a more accurate alternative to standardized testing cycles.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• exposure measurement