Overvoltage and its accidents are common

which seriously threatens the stability and reliability of power system.It is necessary for power grid to obtain accurate

high-quality overvoltage data

as is one of the basis for the precisely transient analysis modeling and safe operation of power system.Previously some relevant research has captured a mess of overvoltage data

meanwhile errors or invalid waveforms are incorporated into the overvoltage database

the low quality data results in valueless and inefficient data mining.The method and framework in the field of overvoltage data cleansing is in great need.In this paper

the measured overvoltage data in power station is studied and overall structure of the measured overvoltage massive data cleansing framework is proposed.The discrepancy between error waveform and real overvoltage waveform is studied.A method of automatical cleansing of massive overvoltage based on sparse autoencode

principal component analysis（PCA） and classification based on clustering for search fast density peak（CFSFDP） is proposed.Self-coding and PCA can automatically extract the attenuation and characteristics of measured waveforms in power system.CFSFDP aggregation method is used to isolate invalid and error waveforms effectively.By establishing the scheme of pre-cleansing and fully cleansing

overvoltage data quality is improved gradiently.Overvoltage database cleansing is conducive to deeply mine the generation and propagation characteristics of overvoltage.