**6. Conclusions and Analysis**

Fault data of main transformer lacks. In addition, the fault mode of the main transformer is different from that of other power transformers. This paper proposes an unsupervised mutual information feature selection method to calculate DGA monitoring data of main transformer and output feature selection sequence. Compared with the supervised mutual information feature selection algorithm, the unsupervised mutual information feature selection algorithm is highly correlated with the sequence features output by the supervised feature selection algorithm in feature selection. In the samples, the training samples and test samples were designed by five-fold method based on the appropriate feature set obtained by the unsupervised mutual information feature selection algorithm. The PSO optimized support vector machine model was used to verify the main transformer fault diagnostic, and the diagnosis accuracy was high. This method is suitable for feature extraction in main transformer fault diagnosis. However, the feature extraction method based on unsupervised mutual information is essentially an embedded feature extraction method with some significant advantages and disadvantages at the same time. The redundancy between features in the selected feature set is minimized, and its limitations depend on the evaluation of candidate solutions by the classification algorithm, which is computationally more expensive. Therefore, the offline data set can be used for training and verification in practical application, and the obtained feature set can be used to judge the condition of nuclear power transformers online.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/ 10.3390/su14052700/s1, the experimental data of DGA samples used in Section 5.1 of this paper.

**Author Contributions:** Conceptualization, R.Y.; data collection, J.T.; methodology and the other, W.Y. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Not applicable for studies not involving humans or animals.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data can be obtained in the Supplementary Materials of the paper.

**Conflicts of Interest:** Jun Tao is an employer from CNNC Nuclear Power Operation Management Co., Ltd. The other authors declare that they have no conflict of interest.
