Research on Abnormal Data Identification and Content Prediction of Dissolved Gas in Power Transformer Oil

Using a neural network model to predict the content of dissolved gas in oil is an important method to evaluate the operation state of a power transformer, and the data quality is the key factor affecting the prediction accuracy of the neural network model. However, due to the complex operating environment of the transformer, there are inevitably many types of abnormal data in the collected gas data, which leads to the decline of data quality. It seriously affects the prediction accuracy of the neural network model. In addition, whether the parameters of the neural network model match or not is also an important factor affecting its prediction performance. However, the traditional selection of parameters based on artificial experience has shortcomings, such as subjectivity, low efficiency, and expansibility, affecting the model's prediction performance to a certain extent. To solve the above problems by modifying the isolation using nearest neighbor ensemble (iNNE), this paper proposes a modified isolation using nearest neighbor ensemble (MiNNE), which uses the characteristics of MiNNE to comprehensively consider local and global metrics to accurately identify gas anomaly data and effectively improve the data quality. At the same time, the pelican optimization algorithm is improved, and an improved pelican optimization algorithm (IPOA) is proposed, and IPOA is used to optimize the key parameters that affect the prediction accuracy of the neural network model, which effectively overcomes the problems of low prediction accuracy of the model caused by traditional empirical parameter selection and the traditional POA easy to fall into local optimization, and improves the prediction performance of the model. The actual operation data of the power transformer verify the model proposed in this paper. The results show that, compared with other models, this model has achieved the best prediction results in the prediction of seven kinds of characteristic gases, which fully proves the superiority of this model.