鉴于现有电动车能耗预测多基于实验室条件,存在结果过于理想化或预测准确度不足的问题.本文基于北京市 51 路公交车的实车运行数据,分析能耗影响因素,通过时钟循环编码优化时间信息、使用箱线图设置阈值以构造行驶工况、建立基于熵权法的驾驶行为评价体系对驾驶行为与工况状态进行辅助分析,最后,对聚类后的 4 类典型工况片段分别建立引入注意力机制的LSTM能耗预测模型,并将其与传统LSTM及LGBM等多种预测模型进行对比分析,验证结果表明引入注意力机制的LSTM预测模型性能显著高于其他模型.
In view of the most of the existing energy consumption prediction of electric vehicles based on the laboratory conditions,the results are too ideal and the actual deviation is large or the accuracy is insufficient.According to the actual running data of Beijing No.51 bus,the influencing factors of energy consumption is analyzed,the time information through clock cycle coding is optimized,and a driving behavior evaluation system is established based on the entropy weight method for auxiliary analysis of driving behavior and operating conditions by using the boxplots to set thresholds to construct driving conditions.Finally,the LSTM energy consumption prediction model by introducing the attention mechanism is established for the four types of typical working condition segments after clustering,and it is compared and analyzed with the various prediction models such as traditional LSTM and LGBM,the validation results show that the performance of the LSTM prediction model by incorporating the attention mechanism is significantly higher than the other models.