Partial Discharge Recognition of Medium Voltage Switchgear Based on CatBoost Model

Medium voltage switchgear is a part of the power distribution system. During the operation and maintenance of switchgear, partial discharge (PD) detection is usually an important means of diagnosing insulation defects. Therefore, accurate identification of PD signals under different types of insulation defects is essential. Previous algorithms in the ensemble learning gradient boosting framework have problems, such as overfitting and low recognition accuracy. This paper introduces the use of Categorical Boosting (CatBoost) algorithm, to identify PD signals in switchgear. The algorithm is the improvement of Gradient Boosting Tree (GBT) algorithm with strong generalization ability. Statistical features of PD signals under different insulation defects are used as inputs to train the model. The result shows that the CatBoost algorithm can efficiently identify PD signals, and evaluate the insulation performance of the switchgear.