汽车燃油消耗预测模型的研究进展
关鹏1,2，任烁今2，沈义涛1，3*，赵健福2
1.哈尔滨工业大学（威海）汽车工程学院，山东威海264209；
2.中汽研汽车检验中心（天津）有限公司，天津300300；
3.上海交通大学动力机械与工程教育部重点实验室，上海200240
摘要：为准确预测车辆在不同工况下的油耗，帮助研发人员深入了解发动机的油耗变化规律并进一步优化发动机性能，全面总结和分析现有的油耗预测模型，将其分为传统油耗预测模型和基于数据驱动的机器学习油耗预测模型两大类，并将基于数据驱动的机器学习油耗预测模型进一步细分为多元回归、浅层机器学习、深度学习和混合式油耗预测模型四类，分析各种模型的预测方法及其变体的应用现状、优势与局限性。明确各类模型的最佳应用场景，指出当前研究中存在的主要问题和挑战。数据驱动的机器学习类预测模型中，多元回归方法对线性相关性强的数据具有很好的表现，且模型透明度高，易于理解；机器学习能够有效应对复杂的非线性关系，特别是深度学习，可以充分挖掘数据中的特征，实现燃油消耗的精准预测，但对数据质量要求较高，模型相对复杂。根据不同模型的特点与应用，对油耗预测未来的发展进行展望。
关键词：油耗预测；数据驱动；机器学习；深度学习；混合油耗模型
Research progress of vehicle fuel consumption prediction models
GUAN Peng1,2, REN Shuojin2, SHEN Yitao1,3*, ZHAO Jianfu2
1.School of Automotive Engineering, Harbin Institute of Technology (Weihai), Weihai 264209, China;
2.CATARC Automotive Research and Inspection Center(Tianjin) Co., Ltd., Tianjin 300300, China;
3.Key Laboratory for Power Machinery and Engineering of Ministry of Education, Shanghai Jiao Tong University, Shanghai 200240, China
Abstract: To accurately predict the fuel consumption characteristics of vehicles under various operating conditions, assist researchers in gaining a deeper understanding of the patterns of fuel consumption changes in engines and further optimize engine performance, this paper aims to comprehensively summarize and analyze existing fuel consumption prediction models and categorize them into two main types: traditional fuel consumption prediction models and data-driven machine learning-based fuel consumption prediction models. For the latter category, this paper further divides it into four subcategories: multiple regression, shallow machine learning, deep learning, and hybrid fuel consumption models, detailing the application status, advantages, and limitations of each method and its variants. Through comparative analysis of these models, this paper not only clarifies their optimal application scenarios but also pointes out the main problems and challenges present in current research. When dealing with data that exhibits strong linear correlations, multiple regression methods perform well, offering high model transparency and ease of understanding. Machine learning approaches, especially deep learning, can effectively address more complex nonlinear relationships, fully exploiting features within the data to achieve precise predictions of fuel consumption, albeit with higher requirements for data quality and relatively complex models. Finally, based on the characteristics and applications of different models, this paper provides an outlook on the future development of fuel consumption prediction.
Keywords: fuel consumption prediction; data-driven; machine learning; deep learning; hybrid fuel consumption model
