Data-driven multi-scenario driving cycle construction for off-road unmanned ground vehicles

High-fidelity driving cycles serve as the fundamental basis for the energy efficiency assessment and powertrain optimization of Off-road Unmanned Ground Vehicles (OUGVs). To address the modeling challenges posed by the strong nonlinearity and high stochasticity inherent in off-road driving processes, this paper proposes a data-driven framework for constructing multi-scenario driving cycles. Based on over 200 h of real-world vehicle data, Kernel Principal Component Analysis is first applied for nonlinear dimensionality reduction, and then Self-Organizing Map and K-means clustering are employed to classify the driving data into four typical exploration task scenarios. Subsequently, utilizing the state transition probabilities of the Markov Chain Monte Carlo process as underlying physical constraints, these probabilities are directly embedded into the crossover and mutation operators of a Genetic Algorithm, accompanied by a specific optimization for standstill phases. Validation results demonstrate that the constructed multi-scenario driving cycles achieve high global statistical fidelity with a mean error of only 4.5%. Furthermore, the framework exhibits remarkably low variability across runs and excellent statistical robustness over multiple independent executions. This study provides a solid data foundation for the parameter matching and precise energy management of unmanned vehicles operating in complex off-road environments.