This paper introduces a new volatility measure and constructs its model based on multifractal volatility method． Taking 5-minute high frequency data of the Shanghai Composite Index as an example，and applying the out-of-sample rolling time window forecasting combined with Model Confidence Set which is proved superior to SPA test，this paper compares the empirical performance of the new model and those of the GAＲCH-type and Ｒealized volatility (RV) models． The empirical results show that the forecasting accuracy of the multifractal volatility measure model in the short term as well as in the long term are better than the GARCH-type and RV models． Moreover，the forecasting models in the long term perform better than those in the short term．The performance in most loss function of the new method based on multifractal volatility measure is superior to other forecasting models．