Probabilistic solar irradiance forecasting based on XGBoost

Solar energy has received increasing attention as renewable clean energy in recent years. Power grid operators and researchers widely value probabilistic solar irradiance forecasting because it can provide uncertainty measurement for future PV production. This paper proposes a probabilistic prediction model of solar irradiance based on XGBoost. Specifically, after data preprocessing, historical data is utilized for training a point prediction model based on XGBoost. Since XGBoost is obtained by minimizing the residuals of successive iterations of multiple trees, when predicting solar irradiance at a certain time in the future, these trees can generate multiple predicted values of irradiance iteratively. Finally, the kernel density estimation method is applied to transform the above prediction results in probability prediction intervals under different confidence levels. Experimental results on public data sets show that this method has better accuracy than other benchmark algorithms. The experiment also shows that the method proposed in this paper requires less training time and simple parameter adjustment, which is very suitable for application in engineering practice. (c) 2022 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the scientific committee of the 2021 The 2nd International Conference on Power Engineering, ICPE, 2021.

Automated summary

The paper presents a probabilistic prediction model of solar irradiance based on XGBoost, which utilizes historical data to train a point prediction model and generates probability prediction intervals under different confidence levels using kernel density estimation. Experimental results demonstrate that this method has better accuracy and is suitable for engineering practice.