Transformer DGA Fault Prediction Based on WOA and T-SNE Optimized Random Forest Algorithm

Dissolved gas analysis (DGA) of transformer oil is an effective means to timely find potential faults in transformers. In order to further improve the accuracy and the applicability of transformer fault prediction model, a new method based on WOA and T-SNE is proposed to optimize the performance of random forest decision-making model. Based on T-SNE , the original DGA sample data of transformer is condensed and features are optimized. The feature vector is reduced from 12 dimensions to 3 dimensions, and the effect of division in sample space is significantly enhanced. Taking the highest prediction accuracy of the random forest model as the objective function, the WOA is used to find the optimal solution of three key parameters in random forest, including the number of trees, the number of split point sand the depth of trees. The DGA fault prediction model of transformer is constructed based on the optimized random forest. Finally, a transformer fault prediction in a substation is taken as an example to verify the consistency and effectiveness of the method with engineering practice. The model is also compared with traditional random forest, professional mechanism model and a variety of machine learning models. The results show that the accuracy of the transformer’s DGA fault prediction model based on optimal random forest is 95.2%, which is 6.7% higher than the original random forest, Compared with other models, it also has higher prediction accuracy and calculation efficiency.