Adaptive Energy Management for Real Driving Conditions via Transfer Reinforcement Learning

  • 2020-07-24 15:06:23
  • Teng Liu, Wenhao Tan, Xiaolin Tang, Jiaxin Chen, Dongpu Cao
  • 1

Abstract

This article proposes a transfer reinforcement learning (RL) based adaptiveenergy managing approach for a hybrid electric vehicle (HEV) with paralleltopology. This approach is bi-level. The up-level characterizes how totransform the Q-value tables in the RL framework via driving cycletransformation (DCT). Especially, transition probability matrices (TPMs) ofpower request are computed for different cycles, and induced matrix norm (IMN)is employed as a critical criterion to identify the transformation differencesand to determine the alteration of the control strategy. The lower-leveldetermines how to set the corresponding control strategies with the transformedQ-value tables and TPMs by using model-free reinforcement learning (RL)algorithm. Numerical tests illustrate that the transferred performance can betuned by IMN value and the transfer RL controller could receive a higher fueleconomy. The comparison demonstrates that the proposed strategy exceeds theconventional RL approach in both calculation speed and control performance.

 

Quick Read (beta)

loading the full paper ...