In power market, accurate electricity price forecasting can help all market participants make optimal bidding or purchasing decisions and maximize their revenue. In recent years, much attention has been focused on the short-term electricity price forecasting. Based on the theory of ARMA-GARCH model, the paper divides the constant day series into working day series and holiday series. Then the models on them are established respectively. The results of PJM electric market show the effectiveness of the proposed methodology, which improves the accuracy of forecasting significantly and is suitable for forecasting day market clearing price and spot price and even other forecasting domains based on the time series models.