Development of a Typical Urban Driving Cycle for Battery Electric Vehicles Based on Kernel Principal Component Analysis and Random Forest

Wang, Lu; Ma, Jian; Zhao, Xuan; Li, Xuebo · 2021 · DOAJ

DOI: 10.1109/ACCESS.2021.3052820

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 need for a tailored urban driving cycle for battery electric vehicles (BEVs) to improve the accuracy of energy consumption and driving range forecasts. Existing legislative cycles, such as the New European Driving Cycle (NEDC) and the China Light-Duty Vehicle Test Cycle (CLTC-P), often fail to represent real-world driving behaviors or specific vehicle types like BEVs. To bridge this gap, the authors propose a novel method for developing a typical urban driving cycle using real-world data from Xi’an, China, employing Kernel Principal Component Analysis (KPCA) and Random Forest (RF) algorithms. The study combined chase car and on-board measurement methods to collect real driving data over one week across various traffic periods. Data from 28 drivers using BYD E6 vehicles were recorded via GPS, OBD, and VBOX equipment. After preprocessing to remove sensor noise and unrealistic acceleration values, the data were segmented into 1,414 valid micro-trips defined by stopping fragments. Fourteen characteristic parameters describing these micro-trips were extracted. To handle the nonlinearity and redundancy of these parameters, KPCA was applied for dimensionality reduction, selecting four principal components that captured 91.96% of the variance, outperforming traditional linear PCA. Subsequently, an improved clustering method combining K-means and Random Forest was used to categorize micro-trips into low-speed, medium-speed, and high-speed groups. This hybrid approach optimized the clustering results, which were validated using metrics such as Compactness, Separation, and the Davies-Bouldin Index. The results demonstrate that the proposed KPCA and Random Forest combination effectively clusters micro-trips into distinct driving modes corresponding to congestion, normal, and smooth traffic conditions. The method successfully generated candidate cycles by splicing representative micro-trips. A target cycle was selected based on assessment criteria considering characteristic parameters and speed-acceleration distribution probabilities. Comparative studies revealed that the proposed method yields a cycle with higher typicality and better representation of real driving behavior compared to other legislative cycles. The analysis of discrepancies between the target cycle and existing standards confirmed its effectiveness in capturing the specific kinematic features of BEVs in urban environments. The significance of this work lies in providing a robust, data-driven framework for developing vehicle-specific driving cycles. By addressing the limitations of linear dimensionality reduction and sensitive clustering algorithms, the study offers a more accurate tool for vehicle design, power matching optimization, and energy management strategy development for BEVs. This approach helps mitigate range anxiety by providing more realistic estimates of energy consumption, thereby supporting the broader adoption and efficient operation of new energy vehicles.

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