Microstructure noise and heavy tail phenomena are very common in financial asset return series under the background of high-frequency trading. At the same time, the covariance matrix of financial asset returns is characterized by high dimensionality and sparsity. Based on the pre average method with Huber loss function, a shrinkage estimation of the covariance matrix of financial asset returns in the context of high-frequency financial data is proposed. The simulation results show that the shrinkage estimation performs well. Finally, high-frequency data of Chinese A-share stock market is applied for empirical analysis to explore the investment performance of the estimator on the minimum variance portfolio. The results show that: (1) The pre average method can eliminate the influence of most microstructure noises on covariance matrix estimation; Huber loss function can also weaken the influence of heavy tail on covariance matrix estimation; (2) Shrinkage estimators can better estimate the overall covariance matrix, and have the better performance in the comparison of minimum variance investment strategies.