A Self-adaptive Energy Management Strategy for Plug-in Hybrid Electric Vehicle based on Deep Q Learning - Details

Author：

Zou, Runnan (Zou, Runnan.) | Zou, Yuan (Zou, Yuan.) | Dong, Yanrui (Dong, Yanrui.) | Fan, Likang (Fan, Likang.)

Indexed by：

EI Scopus

Abstract：

With　the　development　of　energy　management,　deep　learning-based　algorithm　has　become　a　widely　concerned　strategy.　The　presetting　of　neural　network　is　deemed　as　a　key　of　effectiveness　of　the　method.　For　the　purpose　of　improving　fuel　economy　of　plug-in　hybrid　electric　vehicle　(PHEV)　based　on　the　deep　Q　learning,　an　self-adaptive　energy　management　strategy　is　proposed　in　this　paper.　In　order　to　obtain　an　optimal　learning　rate　which　is　one　of　the　key　hyper　parameter　for　deep　Q　network,　deep　Q　learning　(DQL)　with　normalized　advantage　function　(NAF)　and　genetic　algorithm　(GA)　is　combined　together.　The　improvement　of　optimized　learning　rate　is　verified　by　comparing　optimized　learning　rate　with　different　other　learning　rates.　Simulation　results　proves　the　optimized　learning　rate　achieves　the　best　improves　fuel　economy　of　PHEV　compared　with　other　sets　of　learning　rate.　The　result　indicates　the　effectiveness　of　GA　in　finding　an　optimal　hyper　parameter　and　the　effectiveness　GA-NAF-DQL　in　fuel　saving　in　PHEV.　©　Published　under　licence　by　IOP　Publishing　Ltd.

Keyword：

Reinforcement learning Energy management Fuel economy Genetic algorithms Fuels Deep learning Plug-in hybrid vehicles Learning algorithms

Author Community：

[ 1 ] [Zou, Runnan]National Engineering Lab for Electric Vehicles, School of Mechanical Engineering, Beijing Institute of Technology, Beijing; 100081, China
[ 2 ] [Zou, Yuan]National Engineering Lab for Electric Vehicles, School of Mechanical Engineering, Beijing Institute of Technology, Beijing; 100081, China
[ 3 ] [Dong, Yanrui]Beijing University of Technology, Beijing; 100124, China
[ 4 ] [Fan, Likang]School of Mechanical Engineering, Beijing Institute of Technology, Beijing; 100081, China