The capacity prediction of residential charging stations can provide a reference for its capacity selection and contribute to the carbon peaking and caron neutrality goals. In this regard, a datadriven method for predicting the capacity of residential charging stations is proposed. Firstly, historical capacity data of residential charging stations are collected and preprocessed. Secondly, differentsized time windows are used to slice the data as input features. Finally, a prediction model combining multihead attention mechanism and gated recurrent unit neural network is constructed, and the features are input into the model to achieve accurate prediction of future capacity. The results of the case analysis show that the exponential mean absolute error and exponential root mean square error of the model are 33.19 and 102.14% respectively. Compared to other models, the proposed model significantly improves the prediction accuracy and provides new insights for capacity prediction of residential charging stations.