Short-Term Sag Prediction Approach Based on XGBoost Combining Monotonic Constraints with EMA Features

The sag prediction of overhead ground wire is very important, because excessive sag will reduce the safety margin and endanger the transmission reliability, especially under extreme conditions such as heat wave and icing. To solve this problem, we propose a model that combines Exponential Moving Average (EMA) features and monotonic constraints XGBoost. By fusing multi-source meteorological data and sag monitoring data, sag-related features are extracted after outliers elimination and time alignment. Furthermore, EMA features are introduced to capture short-term fluctuations and time dependence. Monotonic constraints encode the physical prior knowledge of “the higher the temperature, the greater the sag”, which improves the physical interpretability. On the measured data, the model’s coefficient of determination (R²) is increased from 0.709 to 0.879, indicating that the short-term prediction accuracy is significantly improved. The combined application of EMA features and monotonic constraints can maintain the physical consistency and enhance the time learning ability, which provides a feasible scheme for intelligent sag prediction of transmission lines.