Development and validation of driving cycles for pure electric vehicles based on autoencoder and Gaussian mixture model clustering

With the intensification of the global energy crisis and environmental pollution, electric vehicles (EVs) have become a key direction in the automotive industry due to their advantages of low emissions and high energy efficiency. However, the promotion of pure electric vehicles still faces numerous technical challenges, among which the development of driving cycles is a critical factor affecting energy consumption evaluation, range prediction, and control strategy optimization. This study proposes a driving cycle development method that combines an Autoencoder neural network and Gaussian Mixture Model. By collecting and preprocessing real-world road data, vehicle operation characteristics were extracted, and driving data were classified into three typical driving conditions: low-speed, medium-speed, and high-speed, using dimensionality reduction via Autoencoder and clustering via GMM. A pure electric vehicle energy consumption model was built using AVL Cruise to validate the accuracy and representativeness of the developed driving cycles. Results show that the deviation of driving cycle characteristic parameters is less than 5%, and the deviation between simulated and actual energy consumption is less than 10%. This study provides theoretical support for energy consumption evaluation and control strategy optimization of pure electric vehicles, contributing to the high-quality development of the new energy vehicle industry.