Fault diagnosis model of DGA in transformer and reactor oil based on time series feature analysis

When DGA online device is introduced into transformer/reactor monitoring, the traditional online monitoring data will produce some problems such as insufficient accuracy of measurement errors. Considering the error generated during the actual operation of the online monitoring device, this paper proposes a sliding time interval sampling sequence that is different from the conventional point sampling, and analyses the trend characteristics of data fluctuation in the sampling interval and wavelet energy characteristics. Combined with IEC dissolved gas ratio methods, feature extraction is carried out from the online monitoring data of the devices with faults found. The fault library is established, and the particle swarm optimization is used to improve the parameters of the multi-class support vector machine. A fault diagnosis model based on time series feature analysis is proposed, and the traditional IEC oil dissolved gas ratio method and the support vector machine diagnosis method using time point data are compared. The simulation analysis results show that the method proposed in this paper has higher classification accuracy and is more suitable for fault diagnosis of online monitoring data.