PSD-BPA is widely used in the simulation calculation of power systems. However

many man-made data errors easily appear because of the special data form of PSD-BPA that brings great hidden danger to the accuracy and reliability of simulation results. First of all

on the basis of identification steps in bad data identification for leakage reactance parameters of transformer

an improved density-based spatial clustering of application with noise(DBSCAN)algorithm is proposed based on feature similarity by combing the characteristics of parameter data in PSD-BPA power flow data. Secondly

a distributed model based on the outlier coefficient is proposed after extracting typical feature data in each classified data set. On this basis

a distinguishing method based on suspiciousness is proposed by referring to the law of distribution. Last but not least

the effectiveness of the model and method adopted in this paper is verified by actual calculation.