Neural network based models for online SOC estimation of LiFePO4 batteries used in Electric Vehicles

Several battery chemistries such as Lead-acid, Nickel-metal-hydride (Ni-MH), Nickel-cadmium (Ni-Cd) and Lithium-ion (Li-ion) are the predominant power source in Electric Vehicle and Hybrid Electric Vehicles (EVs/HEVs). Li-ion batteries are currently the favoured power sources in this regard, due to their operative advantages over other battery chemistries. However, it is challenging to determine the battery’s current capacity state. Contributing factors to these challenges arise due to battery ageing along with the effects of complex electro-chemical reactions. A critical measure to estimating battery capacity is State of charge (SOC), which identifies when a battery reaches its maximum/minimum potential and therefore remaining capacity. This paper presents a comparative study and performance analysis of different artificial neural network (ANN) based State of charge (SOC) estimation models of Lithiumion batteries. Beneficially the proposed technique does not require knowledge of the battery’s internal parameters or electrochemical/mathematical models. Alternatively, it eliminates the highly skilled knowledge requirement by utilising readily available external parameters, such as the battery’s voltage, charge/discharge current at ambient temperature to estimate SOC. Test results show that proposed ANN models achieve higher accuracy, despite ageing and temperature variation effects are not considered.These ANN based SOC models may be used to analyse the effect of operating conditions and energy demand of EVs.