Decomposition-based wind power forecasting models and their boundary issue: An in-depth review and comprehensive discussion on potential solutions

Recently, numerous forecasting models have been reported in the wind power forecasting field, aiming for reliable integration of renewable energy into the electric grid. Decomposition-based hybrid models have gained significant popularity in recent years. These methods generally disaggregate the original time series data into sub-time-series with better stationarity, and then the target data is predicted based on the sub-series. However, existing studies usually utilize future data during the decomposition process and therefore cannot be appropriately employed for real-world applications, due to the inaccessibility of future data. This problem is usually known as the boundary issue. By ignoring the boundary issue during decomposition, the developed decomposition-based forecasting models will inevitably lead to unrealistically high performance than what is practically achievable. These impractical predictions would compromise the scheduling and control decisions made based on them. In light of this, this study provides an in-depth review of decomposition-based models for wind power forecasting, as well as the existing solutions for resolving the boundary issue. We first categorize decomposition-based models with the consideration of the boundary issue, wherein the treatment of the boundary issue varies over different hybrid model architectures (i.e., direct approach and multi-component approach) and decomposition techniques (i.e., empirical mode decomposition, variational mode decomposition, wavelet transform, singular spectrum analysis and hybrid decomposition). Then, we systematically summarize commonly available boundary issue solutions into three categories, namely algorithm-based solutions, sampling-strategy-based solutions and iteration-based solutions. We also evaluate the strengths and limitations of the existing boundary issue solutions and discuss their applicability to different classification of decomposition-based models for wind power forecasting. This study will provide useful references for a wide range of future studies for developing accurate and practical wind power forecasting models.