随着新能源汽车的发展，为实现我国的“双碳”战略目标，减少碳排放，改善空气质量，电动公交车占城市公交的比重越来越大，走在新能源汽车应用的前列。新能源汽车国家监测与管理平台的建立，有效确立了新能源汽车的数据制度，为开展基于车辆大数据的研究奠定了基础。本文在电动公交车实际运营大数据的基础上，对电动公交车大数据进行了分析与处理，开展了电池充放电能量效率的研究，并对电动公交车续驶里程进行了预测。电动公交车实际运营大数据存在杂乱、重复、模糊等问题，在使用原始数据开展研究前，需要对数据进行预处理。本文根据数据预处理的方法，在数据分析的基础上，进行了数据清洗和数据变换，避免了重复数据、缺失数据和异常数据对后续研究的影响。以往针对电动汽车电池充放电能量效率的研究，主要在实验环境中进行，没有考虑电动汽车实际运行的复杂环境。本文在电动公交车实际运营大数据的基础上，开展了电动公交车电池充放电能量效率的研究。根据电动公交车充放电SOC规律，提取了满足电池充放电能量效率的充放电过程；基于移动平均法，计算了电池充放电能量效率。采用最大信息系数算法，从时间、温度维度，对电池充放电能量效率的影响因素进行了研究；确定了电池循环次数、温度，对电池充放电能量效率有一定影响，车辆行驶状况是影响电池充放电能量效率的主要因素。电动汽车准确的续驶里程，可以缓解驾驶员的里程焦虑，为车主制定合理的驾驶规划提供可靠依据。本文在电池充放电能量效率研究的基础上，开展了电动公交车续驶里程的预测。通过对电动公交车SOC和累计里程之间规律的分析，明确了通过预测单位SOC行驶里程，来预测车辆续驶里程的研究思路；结合电池充放电能量效率的研究，构建了车辆行驶状况、电池循环次数、温度等特征。使用RFECV特征选择算法，提取了有效特征；并运用多种模型，对不同比例训练集，开展了单位SOC行驶里程的预测。实验证明，未调参时，LightGBM模型的预测效果最好，调参后，XGBoost模型预测效果最佳；模型最佳拟合度为0.952268，预测效果较好。

With the development of new energy vehicles, in order to achieve the carbon peaking and carbon neutrality goals of our country, reduce carbon emissions and improve air quality, the proportion of electric buses in the urban public transport is growing, and walking in the forefront of the application of new energy vehicles. The establishment of the national monitoring and management platform for new energy vehicles had effectively established the data system of new energy vehicles and laid the foundation for the researches based on vehicle big data.Based on the actual operation big data of electric buses, this paper analyzed and processed the big data of electric buses, carried out the research on the charge and discharge energy efficiency of battery, and predicted the driving range of electric buses.The big data of the actual operation of electric buses has some problems, such as clutter, repetition and ambiguity. Before using the original data to carry out the research, it is necessary to pre-process the data of electric buses. In this paper, according to the methods of data preprocessing, data cleansing and data transformation were carried out on the basis of data analysis, in order to avoid the influence of duplicate data, missing data and abnormal data on the subsequent research.Previous studies on the charge and discharge energy efficiency of electric vehicle battery were mainly conducted in the experimental environment, without considering the complex environment of the actual operation of electric vehicles. Based on the actual operation big data of electric buses, this paper studied the charge and discharge energy efficiency of electric bus battery. According to the SOC charge and discharge law of the electric buses, the charge and discharge processes that satisfying the charge and discharge energy efficiency of battery were extracted. The charge and discharge energy efficiency of battery was calculated based on the MA method. The MIC algorithm was used to study the factors that influence the charge and discharge energy efficiency of battery from the dimensions of time and temperature. This paper determined that the number of battery cycles and the temperature have certain impacts on the charge and discharge energy efficiency of battery, and vehicle driving condition is the main factor that affect the charge and discharge energy efficiency of battery.The accurate driving range of electric vehicles can alleviate the driver‘s range anxiety, and provide a reliable basis for the owner to make a reasonable driving plan. Based on the research of the charge and discharge energy efficiency of battery, this paper carried out the driving range prediction of electric buses. By analyzing the law between SOC and accumulated mileage of the electric buses, the research idea of predicting vehicle driving range by predicting unit SOC driving range was clarified. Based on the research of the charge and discharge energy efficiency of battery, the characteristics of vehicle driving condition, number of battery cycles and temperature were constructed. The RFECV feature selection algorithm was used to extract effective features. In addition, a variety of models were used to predict unit SOC driving range for training sets of different proportions. The experiments showed that the prediction effect of the LightGBM model was the best when the parameters were not adjusted, and the prediction effect of the XGBoost model was the best after the parameters were adjusted. The best goodness-of-fit of the model was 0.952268, and the prediction model works well.