Research on Prediction Model of Dissolved Gas Concentration in Transformer Oil Based on Deep Learning Algorithm

The dissolved gas content in a transformer is a key index to measure its insulation condition. By accurately predicting the dissolved gas content, the early diagnosis of potential faults can be realized. This method can make use of multiple diagnostic indices from the same signal to uniformly characterize each characteristic parameter, so that more accurate prediction results can be obtained. In this project, the non-negative matrix method (NMF) and the improved Extreme Learning machine (ELM) are combined to establish a set of mathematical models suitable for the prediction of dissolved gas content in transformers. A neural network model based on non-negative matrix transformation is proposed and applied to extreme learning machine. The method takes the low-dimensional matrix as the training sample and removes the redundant information, thus improving the accuracy of prediction. The result shows that the algorithm can reduce the dimension of the sampled data and improve the accuracy of the forecast. It can solve the problem of predicting dissolved gas concentration in transformer oil well.