【摘要】 新能源汽车在低温环境下续航里程显著衰减，因此精准预测剩余里程对缓解用户里程焦虑、优化能源管理及提升电池性能都具有重要意义。本文首先基于2023和2024年各自1—4月的30辆新能源车真实行驶数据，开展了数据清洗、特征提取及模型构建。数据清理主要针对时间乱序、缺失值及异常值问题，采用时间序列重排、插值修复及分段筛选等方法，结合荷电状态（SOC）阈值分段策略，显著扩充有效样本量；特征提取主要是结合空气阻力模型及低温特性，从驾驶行为、电池状态、车辆静态参数等多个维度共提取22项特征，经相关性分析最终筛选14项核心特征，引入车辆额定能量与标称续航里程，以增强模型表征能力。在模型构建上，基于提升树算法的多车数据融合模型，通过分层建模策略优化短续航车辆预测，进一步结合增量学习方法，动态更新模型以适应电池性能衰退等时变特性。结果表明，最终模型在长续航与短续航车辆上的预测均方根误差分别为24.483 km和6.425 km，显著优于传统单车模型。本文所提出的预测方法不仅提升了低温条件下的续航预测精度，也为新能源汽车的电池管理与能量优化提供了新的技术思路。更多还原

【Abstract】 The significant range degradation of new energy vehicles(NEVs) in low-temperature environments necessitates accurate remaining mileage prediction to alleviate range anxiety, optimize energy management strategies, and enhance battery performance. This research analyzes real-world driving data from 30 NEVs collected between January and March in 2023 and 2024, implementing systematic data cleaning, feature engineering, and model development. The data processing phase addressed temporal inconsistencies, missing values, and outliers through time-series reorganization, interpolation techniques, and segment filtering. A state-ofcharge(SOC) threshold-based segmentation strategy substantially increased the effective sample size. In feature extraction, 22 characteristics were derived from driving behavior patterns, battery conditions, and static vehicle parameters, with 14 core features selected through correlation analysis. The incorporation of nominal battery energy and driving range enhanced model representation capabilities. The model development utilized a multi-vehicle data fusion approach based on the Boosting Tree algorithm as the foundation. A hierarchical modeling strategy improved predictions for short-range vehicles, while incremental learning enabled dynamic model updates to accommodate time-varying factors such as battery degradation. Experimental validation demonstrated root mean square errors(RMSE) of 24.483 km for long-range vehicles and 6.425 km for short-range vehicles, markedly superior to conventional single-vehicle models. This methodology not only enhances prediction accuracy under low-temperature conditions but also presents novel technical approaches for NEV battery management and energy optimization.更多还原