Challenges in applying the empirical mode decomposition based hybrid algorithm for forecasting renewable wind/solar in practical cases

The hybrid forecasting algorithm, based on empirical mode decomposition (EMD), has attracted considerable attentions and been widely applied to forecast electricity load, wind speed, and solar irradiation time series (TS). The basic idea of the EMD based method is to decompose the complicated original TS into a collection of sub-series and build specific forecasting models for individual sub-series. Final forecasting results of the original TS are obtained by adding up forecasting results of individual sub-series. However, the traditional EMD based forecasting algorithm presents two challenges, which have not been thoroughly discussed in literature and could impede its effective application on practical cases: (1) Decomposed sub-series are very sensitive to the original TS. That is, sub-series with newly obtained TS data may be significantly different from the one used in training the forecasting model. In turn, the established model may not suit for newly decomposed sub-series and has to be trained more frequently; and (2) Key environmental factors usually play a key role in improving wind/solar forecasting results via non-decomposition based methods. However, it is difficult to incorporate key environmental factors into forecasting models of individual sub-series as they do not present strong correlations as compared to the original TS data. This paper gives an in-depth analysis on the EMD based forecasting algorithm, and presents numerical case studies to show its challenge when applying to wind/solar forecasts in practical cases and possible alternatives to solve the challenge. It is expected that this research could bring more attentions to improved algorithms for solving the challenge of the EMD based forecasting method and make it more suitable for practical cases.