With the development of artificial intelligence algorithms

monitoring data of power equipment is widely used in early fault diagnosis of equipment. Since the original monitoring data may contain some abnormal ‘dirty’ data

which will seriously affect the accuracy of intelligent diagnosis of fault of equipment

it is therefore necessary to clean the equipment fault database. In this paper

the linear model and the proximity model are combined to construct a data cleaning method based on ISOMAP equidistant map clustering. Firstly

the ISOMAP analysis method is used to extract the features of the original data and

on this basis

the DBSCAN density clustering analysis is performed. This method fully considers the nonlinear relationship and clustering characteristics between the characteristic quantities of power equipment

removes outliers and improves data quality. In this paper the cleaning of the fault database of the dissolved gas concentration in the oil of transformer is taken as an example. The results show that the diagnosis accuracy rate by fault diagnosis learning network trained by the cleaned database is improved by 17%. The classification and recognition rate of the method proposed in this paper

compared with other data cleaning algorithm models

is improved by 5% and the cleaning result is perfect.