Forecasting of Solar Energy Generation via Dynamic Model Ensemble

Solar energy is one of the primary renewable energy sources. The usage of solar energy generation can help reduce carbon emissions and thus help address the challenge of global climate change. In the power system, we would like to maintain the balance between power generation and power usage. How-ever, solar energy generation is quite intermittent and will be affected by many uncontrollable factors, such as temperature, cloud cover, and humidity levels. Hence, the accurate forecasting of solar energy generation is of significant importance for the secure operation of power grids. Different types of methods have been developed for solar power generation forecasting, including statistical methods and machine learning methods. In this work, we propose a hybrid dynamic ensemble framework for solar energy generation forecasting. Specifically, a set of base forecasting learners will be first learned. Then, a set of weights will be dynamically updated for the base learners based on the expectation of the contributions of the base learners. Extensive experiment results on the real-world dataset have been implemented and demonstrated the effectiveness and robustness of our proposed method. The proposed forecasting method consistently outperforms all single-model baselines and the static ensemble model.