To better obtain the current operating state of the battery energy storage system， a state of charge （SOC） evaluation method of the battery energy storage system based on neural network fusion is proposed. First， the advantages and disadvantages of back-propagation （BP）， gated recurrent unit （GRU）， and long and short-term memory （LSTM） neural network algorithms are compared. The calculation time of BP is usually short， while the estimation accuracy of LSTM for temporal data is high. Then the correlation degree between different input parameters and SOC is analyzed by KL divergence， Pearson correlation coefficient， and grey correlation degree. Compared with the LSTM estimation results， the characteristic parameters that have a greater impact on the SOC of the battery energy storage system are selected. Finally， the empirical mode decomposition algorithm is applied to decompose the SOC data into multiple components， and the sample entropy is used to aggregate the components into high and low-frequency bands. BP and LSTM neural network algorithms are used to estimate SOC in different frequency bands. Compared with a single strategy， the proposed method not only improves the estimation accuracy of SOC， but also reduces the calculation time.