基于LSTM-DDPG的车速预测对增程式汽车能量管理研究

打开文本图片集

中图分类号：U461 收稿日期：2025-02-12 DOI： 10.19999/j.cnki.1004-0226.2025.05.001

Study on Speed Prediction for Energy Management of Add-on Vehicles Based on LSTM-DDPG

Tang Jianxing Yang Chao Yue Zhigang Luo Jiaxin Yu Hengbin Sun Guoyang Automotive Technology Co.，Ltd.，Beijing 10260o，China

Abstract：Inordertoenhancetheenergy management eficiencyofanad-onelectricvehicle（EREV），thepaperfirstlyuses a longshort-term memory（LSTM）neuralnetwork topredictthevehiclesped.Basedonthispredictionresult，thepowerdemandrequiredatfuture momentsisfurthercalculatedandcombinedwiththepowerdemandatthecurent moment，andthesedataare fedto getherintothedeepdeterministicplicygadient （DDPG）intellgence.Tisintellgentbodyisesponsibleforeneratingctrolcommands，andsubsequentlysimulationexperimentsareconductedtoverifytheeal-tieresponsivenessoftheproposedcontrolstrategy. TheexperimentalresultsshowthattheLSTM-DDPGenergymanagement strategyproposed inthisstudyreduces theequivalent fuel consumption by 0.613kg ， 0.350kg ，and 0.607kg ，respectively，compared with the DDPG strategy only，the deep Q-network（DQN） strategy，andtheconventionalpower-folowingcontrolstrategyundertheworldheavycommercialvehicletransientcycling（WTVC） operating conditions. In addition， the difference in equivalent fuel consumption is only 0.128kg when compared to the dynamic programming （DP）control strategy，which showsthe advantagesand high eficiencyof this strategy interms ofenergy saving.

Key words： Incremental electric vehicles;Long andshor-term memory neural networks;Deepreinforcement learing;Energy management

1前言

增程式电动汽车（EREV）作为一种新能源汽车技术，正受到越来越多的关注。（剩余8151字）