With the development of the power system, the data in the power system is growing exponentially, and the complexity is increasing. The stable operation of the power system will face greater challenges. Abnormal data can be generated in the power system due to some factors such as sensor failures, cyber-attacks or human error, the appearance of abnormal data poses a certain threat to the stability and reliability of the power system, thus affecting the normal operation of the power system, so the realization of abnormal data recovery of the power system is of great significance. For the recovery of abnormal data, according to the different processing methods, this paper classifies a variety of power system abnormal data recovery methods into three categories: statistics-based abnormal recovery methods, machine learning-based abnormal recovery methods, and deep learning-based abnormal recovery methods. Then, to address the shortcomings of the currently existing anomalous data recovery methods, the development direction of future technology research is prospected.