Research On Noise Reduction Processing Method Of Power Transformer Condition Monitoring Data

As one of the key components in the entire power grid,the power transformer's normal and stable operation determines whether the entire power supply system can operate smoothly or not.At the same time,due to the continuous enrichment of information collection,communication and storage equipment of power systems,the state monitoring data of power transformers gradually exhibits large-volume,high-dimensional and multi-noise data characteristics.Based on this huge data resource,current researchers focusing more on how to use these data to better evaluate the various indicators of power transformers,while less attention is paid to the data quality of condition monitoring data.To begin with,this research introduced and summarized the classical data cleaning theory model;discuss the data science concept from the theoretical level and discuss the data characteristics of the power transformer condition monitoring data;discuss the new problem of data cleaning in power transformer condition monitoring data under the background of large volume data.Then,Systematically analyze explain the structure and characteristics of the noise reduction self-encoder model,discuss the feasibility and effectiveness of noise reduction self-encoder for data denoising,aiming at the problem of noise reduction processing of power transformer condition monitoring data,based on the gas data in power transformer oil,a stack noise reduction self-encoder model based on deep learning theory is proposed and applied to the normal state of power transformer,the bootstrap method is used to determine the threshold range of reconstructed data,and applied to the data cleaning of data.Finally,aiming at the noise classification problem after data cleaning,a classification model based on support vector machine is established to realize the abnormal data classification function of power transformer,finally complete power transformer data cleaning and noise classification model.Through the corresponding simulation experiments,it is demonstrated that the method has certain data cleaning and noise classification effects on the power transformer condition monitoring data.