Abstract：The state of charge（SOC）of batteries is one of the most important parameters in lithium-ion

battery management technology，and high-precision SOC estimation is beneficial for the grid connection and

control of energy storage stations. Battery charge and discharge data are not only time-series in nature，but also

have certain spatial relationships between feature variables. To improve the accuracy and generality of the

estimation method，a SOC estimation method was proposed for lithium-ion batteries based on a joint convolutional

neural networks-long short term memory networks（CNN-LSTM） network structure. Firstly，the feature

relationships between different dimensions of lithium-ion battery data were obtained through CNN feature

extraction，and then the time series relationships were extracted through the LSTM network structure. The joint

network fully captures the spatial and temporal characteristics of the battery dataset. The experimental results show

that the average error of predicting battery SOC based on the CNN-LSTM joint network model is controlled at

0.65%，which is about 4.4% lower than the average error predicted by a single CNN network and about 0.2% lower

than the average error predicted by a single LSTM network. It has good application prospects.

      Key words：lithium-ion battery；battery state of charge（SOC）；convolutional neural networks（CNN）；long

short term memory networks（LSTM）