State of Charge Estimation of Zinc-Bromine Flow Batteries Using Physics-Informed Long Short-Term Memory Networks

In order to improve the state of charge (SOC) estimation accuracy of zinc-bromine flow batteries during specific discharge phases and address the issue of large SOC estimation errors caused by rapid voltage drops and current fluctuations, this study proposes a physics-informed neural network-based SOC estimation algorithm. The proposed method integrates a first-order resistor-capacitor (RC) equivalent circuit model with a long short-term memory (LSTM) neural network, incorporating the parameters of the equivalent circuit model as part of the neural network parameters. Additionally, physical constraints are embedded into the loss function during network training, ensuring that the training process adheres to physical laws. The estimation results of the physics-informed LSTM (PILSTM) network achieve an average error of 0.90% and a mean squared error (MSE) of 1.52%, demonstrating that the proposed algorithm enables high-accuracy real-time SOC estimation for zinc-bromine flow batteries.