The Lithium-ion batteries are widely used in various electronic products. The state of charge (SOC) estimation is crucial to ensure the safety and extend the lifespan of lithium-ion batteries. The traditional methods have limitations in SOC estimation of lithium-ion batteries due to degradation mechanisms and electrochemical processes. Compared to traditional methods, the deep learning model provides better performance on the basis of the historical data on lithium-ion batteries. The commonly used models include the Gated Recurrent Unit(GRU), the Convolutional Neural Network(CNN) and the Transformer. GRU has a memory function. Transformer excels in processing long time series data and CNN performs well in capturing features. The research proposes a deep learning hybrid model based on the models of Transformer, GRU and CNN for SOC estimation. The causal CNN models and separated features are also used in the hybrid model. The results indicate that the proposed model displays the best performance.