Accurate estimation of the State of Charge (SOC) is key to ensuring power batteries’ safe and stable operation. Due to the complex nonlinear relationship between the parameters of power batteries and SOC, existing SOC estimation methods face difficulties in effectively characterizing long-sequence data dependencies. Therefore, a BiLSTM-MHSA method is proposed for the SOC estimation of power batteries. This method takes the current, voltage, and temperature of the power battery as inputs to the BiLSTM model, and uses a gated unit with a memory function to select and forget historical data features. The BiLSTM while processing long-sequence data, treats all data features equally, making it less effective at representing data features that significantly impact SOC. The MHSA unit, however, can determine the focus of feature attention based on the characteristics of the data, improving the model's ability to extract features both globally and locally. This improves BiLSTM’s limitation in paying insufficient attention to historical data features and enhances the model’s estimation accuracy. Finally, comparison and generalization experiments were designed, and the results show that the root mean square error of the proposed method is reduced by 0.63% compared to BiLSTM on a public dataset, verifying that the method can effectively represent long-sequence data features under the influence of complex nonlinear relationships, thereby improving the accuracy of SOC estimation.