Heavy Overload Forecasting of Distribution Transformers Based on XGBoost

As the last-level power supply unit for low-voltage users, the distribution transformer directly affects the power quality of the station. When using big data to predict the heavy overload risk of distribution transformers, it faces the problem of class imbalance. This paper considers the sample class imbalance problem from the data and algorithm level, and built a heavy load risk prediction model for distribution transformer based on Extreme Gradient boost (XGBoost). The model used a serial integrated multi-regression tree to score the heavy overload risk of the transformer and the feature importance ranking is calculated. By comparing logistic regression and XGBoost to predict the heavy overload risk of a prefecture-level city, XGBoost has obvious advantages in terms of calculation speed and prediction accuracy, and makes the model interpretability enhanced.