In order to improve the accuracy of short-term load prediction and provide a more powerful guarantee for stable operation of power system, a short-term load prediction method that applies temporal convolutional network (TCN) and gated recurrent unit (GRU) is proposed in this paper. Firstly, training data are divided into two types, namely time series data and non-time series data. Secondly, time series data are selected as the input of TCN model to extract time series features. Then, time series features and not-time series date are input in the GRU model for training. Finally, the trained model is used to predict short-term power load. Based on real load data of an industry in Foshan City, Guangdong Province, the load forecasting ability of TCN-GRU model is verified. By comparing with the prediction effects of other four deep learning models, the proposed model in this paper is verified to have the ability of forecasting much more accurately for short-term load.