Real-Time Optimal Energy Management of Electrified Powertrains with Reinforcement Learning

被引：18

作者：

Biswas, Atriya ^{[1
]}

Anselma, Pier G. ^{[2
]}

Emadi, Ali ^{[1
]}

机构：

[1] McMaster Univ, MARC, Hamilton, ON, Canada

[2] Politecn Torino, Dept Mech & Aerosp Engn DIMEAS, Turin, Italy

来源：

2019 IEEE TRANSPORTATION ELECTRIFICATION CONFERENCE AND EXPO (ITEC) | 2019年

关键词：

Automotive systems; electric and hybrid electric vehicles; electrified powertrains; energy management system; premeditated EMS; Q-learning; real-world driving scenario; real-time; reinforcement learning;

D O I：

10.1109/itec.2019.8790482

中图分类号：

TM [电工技术]; TN [电子技术、通信技术];

学科分类号：

0808 ; 0809 ;

摘要：

Reinforcement learning (RL) algorithm is employed in solving energy management problem for electrified powertrain in real-world driving scenarios and the application process is streamlined. A near-global optimal control policy is articulated for the energy management system (EMS) using Q-learning algorithm which is real-time implementable. The core of the EMS is an updating optimal control policy in the form of a changing look-up table comprising near-global optimal action value function (Q-values) corresponding to all feasible state-action combinations. Using the updating control policy, the EMS can optimally decide power-split between electric machines (EMs) and internal combustion engine (ICE) in real-world driving situations.

引用

页数：6

共 15 条

[1] Explicit optimal control policy and its practical application for hybrid electric powertrains [J].