Abstract: In the traditional short-term load forecasting, the coupling effect of real-time meteorological factors is not considered.To solve the above problem, a short-term load forecasting method of temporal convolutional network is proposed considering the real-time meteorological coupling effect. First, the correlation between the real-time comprehensive meteorological indices and load curves is analyzed, and then the similar day selection method of the mixed daily characteristics and the real-time meteorological factors is constructed. Moreover, the real-time comprehensive meteorological indices are introduced as the model inputs. Finally, the temporal convolutional network which can fully consider and contain the characteristics of"time difference"between the real-time meteorological factors and the loads is used to carry out the day-ahead load forecasting modeling. The experiment simulation takes the actual load of a regional power grid as an example, and the simulation results show that the forecasting model can effectively improve the accuracy of day-ahead load forecasting in regional power grid.