For non-stationary short-term household load data, it is difficult to mine deeper temporal characteristics by directly applying the prediction model. A combination of Empirical Mode Decomposition (EMD) and Stack Long Short-Term Memory (SLSTM) algorithm is proposed for short-term household load forecasting. Firstly, the principle of EMD and SLSTM is analyzed and the EMD-SLSTM combined prediction model is proposed. Then, the load data is decomposed by the EMD algorithm and the decomposed component data is respectively converted into three-dimensional data. By designing the network architecture of SLSTM and its parameters, the normalized component data and original data are separately predicted and reconstructed. In order to show the performance of the algorithm, the performance of the support vector regression, artificial neural network, deep neural network, gradient boosting regression is compared and verified by MAPE and RMSE performance metrics in two scenarios. The results show that EMD-SLSTM can more effectively express the time series relationship of short-term household load and has higher prediction accuracy. This work is supported by Youth Science Foundation (No. 51607179).