Based on the existence of both realized volatility measurement errors and sudden changes in market structure in real financial markets, the Markov regimeswitching is introduced into the model corrected by errors, and the MRS-HARQtype models with time-varying characteristics are constructed to further improve the forecasting accuracy of volatility models under the HAR framework. The empirical study with the sample of 5-minute high-frequency data of the CSI 300 index from April 2005 to July 2021 confirms that, compared with the HAR-type and HARQ(F)-type models, the MRS-HARQ-type models constructed by combining errors and Markov regime-switching have better goodness-of-fit and forecasting performance, and can more accurately portray the complex characteristics of the real volatility of China's stock market, and finds that the high volatility states do not last long; furthermore among the 16 high-frequency volatility models explored in this paper, the MRSHARQ-RS model combining the realized semivariance and the MRS-HARQ-SJ model combining signed jump possess the best out-of-sample forecasting performance. The findings have good applications in the fields of financial risk modeling, asset pricing and financial forecasting.