Condition-based maintenance of transformers based on L1 regularization

Power transformer is one of the major power supply equipments in the electric power system, whose reliability is directly related to the safe running of power system. So, condition-based maintenance of transformers is very important. Recently, some data mining techniques such as C4.5 decision tree, artificial neural network and SVM have been employed to assist condition- based maintenance tasks for transformers. But the models obtained have no good enough prediction accuracy and satisfactory sparsity. We establish L₁ regularization classification model and propose an improved gradient boosting algorithm based on a cost-sensitive loss function to solve the problem. The numerical results of a real data show that the prediction accuracy of the L₁ regularization model is high enough. Furthermore, the solutions are sparse and easy to be interpreted.