The electric characteristic of microgrid’s battery has obvious nonlinearity and irregularity at work, it is difficult to accurately estimate the using the traditional mathematical methods. According to the problem above, the topology of back propagation (BP) neural network is constructed, and the network model is trained with new adaptive algorithm to improve the traditional learning model, the weights among neurons in the neural network model are adjusted reasonably, and the error is reduced with higher efficiency. The simulation result shows that the estimated results are within the scope of preset accuracy, the average error is less than 4%. It indicates that the BP neural network using the optimized algorithm can accurately estimate the state of charge, the attempt is effective and feasible.