Forecasting dissolved gases content in power transformer oil based on weakening buffer operator and least square support vector machine–Markov

Abstract

The early detection of potential power transformer failures can ensure the safe operation of transformers. So it is practical to develop the early-fault-forecasting technology for transformers. Dissolved gas analysis (DGA) in power transformer is a significant basis for transformer insulation fault diagnosis, which provides full evidence for general internal transformer hidden dangers. But because of the stochastic growth and the small quantity of time-sequence data, forecasting the accurate dissolved gases content in power transformer oil is a complicated problem until now. Least square support vector machine (LSSVM) has been successfully employed to solve regression problem of nonlinearity and small sample. Aiming at improving the primitive shock and disturbance of time-sequence data, this paper firstly introduces the weakening buffer operator to attenuate its randomness. Then, in order to decrease the forecasting error and maximize the total forecasting precision, the Markov chain, which can well reflect the randomness produced by the system involved with many complex factors, is presented to modify the values forecasted by LSSVM. The experimental results indicate that the proposed model can achieve greater forecasting accuracy than GRNN and LSSVM model under the circumstances of small sample.