To overcome the difficulty in electricity consumption forecasting based on meteorological data where observed present meteorological characters are necessary, and improve the accuracy of forecasting, an electricity consumption prediction method based on temporal convolution network (TCN) and recurrent neural network (RNN) was proposed. First, future weather condition is predicted based on historical meteorological data with TCN, and then future electricity consumption is predicted based on future weather condition and historical electricity consumption. During this process, the proposed method depends only on historical data, and without necessity of the observation of present weather condition. Experimental results on real-world meteorological dataset and electricity consumption dataset demonstrated that the proposed method showed an advantage in electricity consumption prediction task when weather is the only external factor used, and of high applicability.