A GRU-RNN based momentum optimized algorithm for SOC estimation

Abstract For a lithium battery, a gated recurrent unit recurrent neural network (GRU-RNN) based momentum gradient method is investigated to estimate its state of charge (SOC). In the momentum gradient method, the current weight change direction takes a compromise of the gradient direction at current instant and at historical time to prevent the oscillation of the weight change and to improve the SOC estimation speed. The details include: (1) construct a GRU-RNN model for estimating SOC by taking the measured voltage and current as the inputs, and the estimated SOC as the output of the GRU-RNN; (2) to promote the SOC convergence speed, explore the momentum gradient algorithm to optimize the weights of the network by introducing a momentum term; (3) to prevent overfitting and to improve generalization ability of the GRU-RNN model, add noises to the sample data, so as to improve the SOC estimation accuracy; (4) set up a lithium battery test platform to sample data in battery discharge process and to implement MATLAB simulation. The simulation results verify that the momentum optimized GRU-RNN model can accurately and effectively estimate the SOC of the lithium battery.