Prediction of dissolved gas in power transformer oil based on random forests algorithm

In order to adopt reasonable measures to anticipate and avoid possible internal failures in power transformers, accurate prediction of power transformer oil dissolved gas trends is useful. It is necessary to realize the condition based maintenance of power transformers, and the prediction of dissolved gas in the oil is solid foundation. The gas concentration is affected by many factors, so prediction model built by reasonable selection of the larger correlation factors will help to improve prediction accuracy. Random forests algorithm is used to predict the gas trends and the prediction performance evaluation is realized by appropriate evaluation indexes. By comparison with support vector machine prediction, the advantage of random forests method for power transformer oil dissolved gas prediction is proved.