In this study, a GM(1,1)-RV model is proposed to forecast the realized volatility. The heterogeneity of investor with different time horizons is taken into account. Unlike previous studies that used linear regression models, GM(1,1) is employed to model weekly or monthly realized volatility trend based on only 5 or 22 data. The empirical results based on Chinese stock market are demonstrated that the GM(1,1)-RV model can generate better forecasting performance than widely used HAR-RV-type regression models from statistical viewpoint and economic perspectives.