Abstract: Phasor measurement units (PMU) have the advantages of good synchronization, high resolution, direct phase angle measurement, etc. It is an important information source for realizing on-line real-time state perception of power systems. However, due to the influence of equipment failure, climate interference, communication problems and other factors, PMU data in the actual power grid are prone to data loss and anomalies, which will interfere with the subsequent advanced power grid applications based on PMU data, thereby affecting the reliability of power grid state perception and operation scheduling. Four kinds of low-quality data are summarized by analyzing the PMU data measured in the field, and the operating state of the system is identified by using mechanism analysis and correlation analysis methods. Then, combining the multi-view learning method with the power grid operation mechanism, a preliminary multi-view data reconstruction algorithm based on spatio-temporal information feature fusion is proposed to reconstruct the low-quality and missing PMU data. Finally, according to the characteristics of different running states of the system, the low quality data are identified by using different views to generate data, and an adaptive weighted missing data reconstruction method based on historical data is proposed. Simulation and measured data show that this method can effectively identify and reconstruct PMU low quality data in real time, which provides effective guarantee for the application of PMU data in power systems.