A method for fast mining abnormal state information of power equipment based on time series analysis

An approach based on time series analysis is proposed as means to mining the abnormal state information of power equipment. The state information is equivalent to the amount of time series of each state. Anomalies in time series are divided into two categories: anomaly caused by intervention, noise and missing values. Then based on the model fitting results, we get the type of anomaly in time series. Considering the noise points and missing value, the iteration method is used to clean data and improve its quality and availability. As to intervention, we introduce the in response structure to infer the existence of latent failures and achieve the early-warning of equipment condition. Finally, the approach is tested the on-line monitoring data of a transformer, with the results showing that this method can effectively identify anomaly and accomplish the goal of data cleaning and state early-warning.