A Driving Cycle Development Method Based on Large-Scale Data and Multi-Dimensional Dynamic Time Warping

Standardized driving cycles often fail to represent real-world driving conditions, limiting their effectiveness for hybrid electric vehicles energy management strategy (EMS) optimization. Current research on customizing driving cycles typically relies on limited datasets. Moreover, existing methods often construct feature spaces using statistical parameters, such as average speed, neglecting the importance of transient features inherent in speed-time profiles, to which EMS is particularly sensitive. In this study, a novel method for driving cycle development is proposed that leverages large-scale real-world driving data compression technology and multi-dimensional Dynamic Time Warping (DTW) to generate representative cycles considering the transient feature. The DTW algorithm is adept at extracting these crucial transient driving features. To address the computational challenge brought by DTW, a Long Short-Term Memory auto-encoder is designed for data compression, achieving a compression ratio of 500 while maintaining a speed–acceleration frequency distribution difference low to 0.827%. After micro-trip classification via principal component and clustering analysis, both one-dimensional and two-dimensional DTW algorithms are employed to extract representative micro-trips. A comprehensive evaluation demonstrates that the driving cycle generated by 2D DTW exhibits superior representativeness compared to a traditional statistical method and 1D DTW method. When applied to EMS optimization, the driving cycle generated using 2D DTW achieves better fuel economy, with up to nearly 3% reduction in equivalent fuel consumption per 100 km. The results demonstrate that the proposed method can effectively generate highly representative driving cycles based on large-scale data and 2D DTW, significantly enhancing the optimization performance of EMS.