HAＲ-type models can be used to describe the proportion of contributions by different class of traders (Heterogeneous) ． They are widely used in empirical study and their performance is good in financial market volatility forecasting． The empirical results show the HAＲ-type models can be used to partly capture the long memory which is very important for financial market，but the performance is very limited． However， AＲFIMA models are very good at describing long memory． Therefore，based on the advantages of above mentioned two models，a new model: HAＲFIMA is proposed and developed further into an HAＲFIMA-type model． Then the HAＲFIMA-type model is used to forecast the realized volatility (ＲV) of S＆P 500 and SSEC． The empirical results show that our HAＲFIMA-type model can be used to capture long memory more accurately，and that the out-sample forecasting performance is better than other models． This conclusion is also robust．