Abstract: Accurate power load forecasting plays an important role in ensuring the stable operation of the power system. In view of the low precision of traditional short-term power load forecasting methods, the sub-sequence fusion problem is not considered after modal decomposition, this paper proposes a multi-scale short-term power load forecasting based on variational mode decomposition (VMD) and temporal convolutional network (TCN) methods. First, VMD is used to decompose the power load data into several sub-components to solve the problems of non-linearity and randomness in the power load data, then TCN is applied to train several components with different time scales, finally a fully connected network is used to analyze each component. Time-scale decomposition signals are fused to realize short-term power load forecasting and improve forecasting accuracy. The experimental results show that the root mean square error (RMSE) is reduced by 40% and the curve fitting is improved by 1.1% compared with the traditional prediction method of VMD and improved long-short-term memory network.