A Seq2Seq prediction method for the state of charge of electric bus batteries was proposed to address the difficulty of predicting the state of charge under high-dimensional input features and long-term prediction needs. The battery state prediction was enhanced by incorporating vehicle driving status and operating conditions. The impact of various factors on the state of charge was analyzed using feature selection algorithms. The WN-Seq2Seq algorithm integrated Seq2Seq and WaveNet cyclic structures, which could enhance the memory and representation ability of high-dimensional input features and predicted state of charge sequence information. The accuracy of the model's prediction was improved. Validation using actual driving data from four electric buses in Hangzhou from 2021 to 2022 demonstrated the effectiveness of the WN-Seq2Seq model. The mean squared error (MSE) reached 0.505%, the mean absolute error (MAE) reached 0.479%, the mean absolute percentage error (MAPE) reached 0.656%, and the model computation time (MCT) reached 0.017 s. The prediction accuracy and stability were significantly promoted compared with traditional models. The WN-Seq2Seq model exhibited good prediction performance across different temperature ranges. The findings provide reliable decision-making support for energy consumption control strategies and safety management of electric buses.