High-accuracy driving cycle test set construction framework for vehicle calibration and measurement

Abstract

The construction of representative driving cycles is crucial for vehicle energy consumption analysis, emission evaluation, and control strategy optimization. However, classical approaches face trade-offs between exploration scope, computational efficiency, and perturbation controllability. This study proposes a novel Markov Chain Divergence-Convergence Domain (MC-DCD) method to address these challenges. The proposed approach first implements reachable domain screening for candidate cycle sub-segments through divergence-convergence domain analysis, ensuring thorough exploration of potential optimal solutions. Subsequently, it achieves minimal perturbation to existing high-quality cycles through rapid reconstruction of local segments in suboptimal cycles. Validated using 1 million seconds of real-world driving data from Qingdao, the MC-DCD method generated 100,000 candidate segments in just 0.66 s. The final driving cycle showed an average statistical deviation of only 0.11%. Compared with classical Markov Chain (MC) and Markov Chain-Genetic Algorithm (MC-GA) methods, the proposed approach improved accuracy by 97-fold and 12.5-fold, respectively, while achieving over 100-fold and 10-fold acceleration in candidate cycle generation speed. Beyond enhancing traditional vehicle evaluation, the massive high-fidelity datasets generated by this method serve as a critical training foundation for neural network-based models, effectively mitigating the challenge of large-scale data scarcity.