Combined Forecasting Method of Dissolved Gases Concentration and Its Application in Condition-Based Maintenance

Abstract

In order to predict the changing trend of the dissolved gas in the transformer oil accurately and guide the condition-based maintenance of a power transformer, a combined forecasting model based on cross-entropy is proposed and its application is analyzed. First, a combined forecasting model is established on the basis of radical basis function neural network, back propagation neural network, least squares support vector machine with two different kernel functions, and grey model. Then, the weight coefficient of each algorithm is determined by cross-entropy theory. After that, an application method of the combined forecasting model by the pattern of a variable time window is proposed for the purpose of arranging the equipment maintenance in advance. In this way, the increased budget due to emergency maintenance can be alleviated, and potential loss due to lack of timely maintenance can be reduced. Finally, the accuracy of the combination forecasting method and the effectiveness of the application method are verified by a series of examples.