Forecasting renewable energy generation based on AutoML and transfer learning

Renewable energy generation is characterized by high volatility and uncertainty. Accurate power forecasting is vital for ensuring the safe operation of the power grid and achieving efficient economic dispatch. In recent years, research on wind and photovoltaic power forecasting has focused on developing complex deep neural network architectures to improve prediction performance. These customized models require extensive expert knowledge and often necessitate designing separate models for different datasets and tasks. In contrast, AutoML aims to automatically select and train models from a pool of generic models, generating highly accurate predictions with limited expertise in the target domain. However, there is currently a lack of research on the application of AutoML in renewable energy generation forecasting, and this study aims to fill this gap. More specifically, we propose an AutoML method with a Pattern-Matching-Based Model Transfer (PMBMT) technique. Unlike previous approaches that assign fixed weights to models, our proposed method utilizes PMBMT to dynamically allocate weights to models by matching new samples with similar historical samples. This allows the trained models to adapt to the latest sample distribution. Experiments conducted on real-world datasets from a wind farm and a photovoltaic plant show that AutoML method can accurately forecast renewable energy generation with minimal expert intervention and our proposed PMBMT can further enhance the accuracy of AutoML.