Data Decomposition Strategy to Improve Solar Forecasting Accuracy

One of the key tasks for the planning, balancing, and management of the grid is an accurate estimate of solar irradiation. The solar global horizontal irradiance (GHI) is intermittent to the climatic characteristics and geographical locations. Therefore, two deep learning (DL) models: (Long short-term memory (LSTM) and Gated recurrent unit (GRU), are developed in this study to forecast accurate and precise solar GHI. Furthermore, to examine the impact of signal processing approach on accuracy, the LSTM and GRU are also used in combination with the complete ensemble empirical mode for decomposition with adaptive noise (CEEMDAN). The CEEMDAN decomposes the dataset to decompose in to several intrinsic mode functions (IMF) and improve the learning ability of LSTM and GRU. As a results, it is observed that the decomposition technique improves the forecasting accuracy by reducing the forecasting errors (RMSE: 31% and MAE: 40%).