Real-Time Prediction of Anode Potential in Li-Ion Batteries Using Long Short-Term Neural Networks for Lithium Plating Prevention

The fast charging technology is urgently needed for wide acceptance of electric vehicles. And the most severe issue in fast charging is the lithium plating due to the low anode potential. In order to prevent lithium plating, it is crucial to monitor the anode potential at different operating conditions. This paper proposes a long short-term memory (LSTM) neural network that predicts the anode potential by using the most commonly measured signals including battery current, voltage, state of charge, and surface temperature. The proposed LSTM neural network is fitted to the training data generated using an experimentally validated battery model. The predictions achieve high accuracy, only 3.84 mV for the maximum Root Mean Square Error (RMSE) on the driving cycles, and 3.73 mV for the RMSE on the constant charging profiles. The results demonstrate the effectiveness of the LSTM neural network in predicting the anode potential. Unlike the existing mathematical models, the data-driven approach used in this paper does not involve complicated mathematical formulation, tedious parameter tuning, or a deep understanding of the electrochemistry. Accurate estimation is accomplished by fitting the neural network to the training dataset. The trained LSTM model is also quite computationally efficient, which enables the real-time estimation.