Abstract: Aiming at the problem that the prediction accuracy of a multi-task learning (MTL) model is limited due to the difference in sensitivity of the meteorological factors to multi-load changes and the difference in coupling intensity between multivariate loads, a MTL and single-task learning (STL)-combined multi-loads forecasting method is proposed. Firstly, the MTL model based on the long and short-term memory (LSTM) network is used to extract the coupling information between multiple loads for preliminary prediction. Then the STL model based on the dual attention before the LSTM (DABLSTM) network is used to reduce the input noises for secondary prediction. The preliminary predicted values are fed into the single-task learning model, allowing the STL model to take future time series information into account. Finally, the prediction results of the two models are fused through the fully connected layer to obtain the final prediction result. The experimental results show that the proposed combined model has higher prediction accuracy compared to the single MTL or the STL model.