Through analysis, it is concluded that there is a correlation between electricity price and load, so the influence of real-time load should be considered in an electricity price forecasting model. We consider the problem that a feedforward neural network can't deal with time series data and the slow forecasting speed of the LSTM neural network. A real-time load forecasting model based on Attention-GRU is proposed. The model makes full use of the time series characteristics of electricity price, and uses an Attention mechanism to highlight the key input characteristics of electricity price forecasting. Taking the real-time data of PJM power market in the United States as an example, it is verified that this method has higher forecasting accuracy and faster forecasting speed than the LSTM neural network by comparing with other forecasting models. This work is supported by Natural Science Foundation of Shaanxi Province (No. 2017JM6105) and Basic Research Plan of Natural Science in Shaanxi Province (No. 2019JLM-11).