Due to the small amount of input information available with load and price selected as input features in electricity price forecasting, it is difficult to achieve a better forecasting effect. In order to accurately capture the change frequency of short-term electricity price, it is proposed to extract the characteristics of electricity price based on the principle of real-time electricity price, followed by an analysis of the causes of electricity price fluctuation from the perspective of electricity price formation mechanism, with the characteristics of real-time electricity price selected for short-term electricity price prediction. Moreover, a recommendation has been made of Seq2Seq-attention network, which is good at capturing the law of electricity price forecast data to forecast. The validity of this proposed method has been verified by the public data of PJM power market in the United States.