关键词: 光伏功率区间预测, XGBoost, 改进雪雁算法, 自助法, SHAP可解释分析

Abstract: Aiming at the current problems such as low prediction accuracy of photovoltaic power and the lack of interpretability of the model, a photovoltaic power interval prediction model based on ISGA-XGBoost and SHAP is proposed. Firstly, the statistics of the sliding time window of photovoltaic power are introduced into the input features to capture the dynamic change trends and patterns of the time series, and a prediction model based on XGBoost is established. Through the regularization strategy and parallel computing optimization, high-dimensional features are processed and overfitting is suppressed. Secondly, ISGA is adopted to integrate the head goose rotation mechanism, the call guidance mechanism and the outlier boundary strategy to improve the hyperparameter optimization ability of the XGBoost model. Then, considering the uncertainty of photovoltaic power, the bootstrap method is adopted to quantify the prediction intervals at different confidence levels. Finally, the SHAP interpretable model is introduced to quantify the contributions of each feature variable and improve the interpretability of the prediction results. The results of the calculation examples show that the model proposed in this paper has higher prediction accuracy compared with other models, and has better generalization ability and interpretability.

Key words: photovoltaic power interval prediction, XGBoost(eXtreme gradient boosting), improved snow geese algorithm(ISGA), bootstrap method, SHAP(Shapley Additive Explanations)