Abstract: On-demand scaling of microservice containers improves resource utilization of clusters, but it requires accurate resource prediction. Current rule-based strategies are prevalent in industry, but academic research has combined machine learning for load prediction, but still have issues like inaccuracy and overfitting. A new model for load prediction based on microservice dependency is proposed. It used a container dependency detection algorithm based on DTW and analyzed the correlation of indicators between containers with strong dependency relationships. The core prediction model used a Seq2Seq encoding-decoding model and combined the attention mechanism and residual LSTM to enhance the stability and accuracy of the model's prediction. The experiment results demonstrate that the proposed model has a significant prediction performance, with an average decrease of 48%, 35% and 51% in MAE, RMSE, and MAPE compared with the other two deep learning models. It effectively forecasts the short-term load of containers with strong dependency relationships.