Using Holt-Winters and IF-XGBoost model to detect the abnormal condition of PV system

Photovoltaic (PV) power generation is one of the primary forms of electricity generation. This paper proposes the Holt-Winters(HW) and IF-XGBoost model based on Bayesian optimization to identify the abnormal condition of PV systems. HW is used to predict the expected output power, and multidimensional features are constructed using the predicted power and time series. The study construct the IF-XGBoost model with isolation forest(IF) and XGBoost, adopting the Bayesian optimization to optimize the hyperparameters in HW and IF-XGBoost. The proposed methods are validated by comparing the accuracy of different models. The result shows that the IF-XGBoost model is better than other models, whose F1-score is 0.981 on the test set. Furthermore, IF-XGBoost demonstrates better performance on the test set with the white Gaussian noise. This paper provides a scheme for early abnormal condition detection and early warning in large-scale PV systems.