Data-Driven Generation and Validation of Synthetic EV Drive Cycles Using GMM and HSMM Techniques

Inhalt:
EV performance assessment can be made by testing the vehicle on the drive cycle. Performance assessment of EV is done through different drive cycles, and the selection of these drive cycles is very crucial for the validation of EV Powertrain and energy storage system. Conventional drive cycles are bound by an inability to produce a dynamic scenario. So, there is a need to generate such drive cycles through advanced data-driven methods, which replicate the highly dynamic scenarios. This paper presents a data-driven framework to generate and validate synthetic electric vehicle (EV) drive cycles using Gaussian Mixture Models (GMM) and Hidden Semi-Markov Models (HSMM). Real-world driving data, including speed, throttle, and brake inputs, is first used to extract relevant time-series features. GMM is applied for clustering driving patterns, and HSMM captures both the temporal dependencies and dwell times of those patterns. The trained model generates synthetic sequences that mimic real driving behavior. These synthetic cycles are reconstructed into full-time-series signals and evaluated using a battery and motor simulation model to analyze torque response and State of Charge (SOC) variations. Various validation techniques, such as statistical comparisons, power spectral density analysis, and cross-correlation with real data, confirm the effectiveness of the proposed method. This framework will significantly create additional resources to augment extensive road testing, offering a reliable and scalable approach for EV testing and design optimization.