Development of Machine Learning Methods in Hybrid Energy Storage Systems in Electric Vehicles

Table 2

Comparison of reinforcement learning algorithms used in HESSs.

Reference	Drawbacks	Advantages	Power transmission system	Algorithm
Hsu et al. [27]	Simplification of complex models	Adaptive to riding conditions	Electric bike	Q-learning
Qi and Wu [28], Liu and Murphey [29]	Dependence on driving data	High accuracy	Hybrid electric vehicle	Temporal-difference (TD) learning
Liu et al. [30]	Sporadic local optimization	Ability to run online	Plug-in hybrid electric vehicle	Q-learning
Hu et al. [31]	Computational load	Multiple control objectives	Hybrid electric vehicle	Q-learning
Kamet et al. [32]	Design complexity	Robust against variability	Hydraulic hybrid vehicle	Deep reinforcement learning and dynamic neural programming
Zhao et al. [33], Xiong et al. [34]	Complex mathematics	Real-time control	Hybrid truck	Dynamic learning
Kamet et al. [32] Zhao et al. [33]	Needs specific training	Data-driven model	Plug-in hybrid electric vehicle	Deep learning
Liu et al. [30], Hu et al. [31]	Sensitivity to driving cycle	Fast computation	Electric vehicle	Online reinforcement learning
Hay et al. [35], Lin et al. [36]	Data requirements	Improved battery life	Hybrid electric vehicle and plug-in hybrid electric vehicle	Reinforcement learning and Markov decision system