A Surface Temperature Estimation Method for Lithium-Ion Battery Using Enhanced GRU-RNN

To monitor the thermal performance of the battery, the surface temperature (ST) of the battery is normally directly measured by temperature sensors. As the number of battery cells or strings increases, the number of temperature sensors increases proportionally. This increases the cost and reduces the reliability of the battery systems. To solve this problem, this article introduces a method to accurately estimate the ST of lithium-ion batteries using a recurrent neural network (RNN) with gated recurrent unit (GRU). First, this article analyzes the battery ST distribution theory and proves that it is a time series task since the present ST is conditioned on the previous state. Second, a GRU-RNN model is adopted to estimate the battery ST as this model has the ability to automatically encode dependencies in time and accurately estimate the battery ST without using any physical battery models or filters. Third, an improved data normalization method is proposed to enhance the estimation accuracy and robustness. Fourthly, the proposed data normalization method is incorporated into the stacked GRU-RNN to estimate the battery ST from compulsory online signals. The proposed method is verified with LiFePO4 using US06 and Federal Urban Driving Schedule (FUDS) profiles under four fixed ambient temperatures and with LiNiCoAIO2 using a mixed dynamic profiles under varying ambient temperature ranges (from 10 °C to 25 °C). The estimation error using mean absolute error (MAE) is less than 0.2 °C over all the fixed ambient temperature conditions and 0.42 °C over the varying ambient temperature conditions.