A novel mode-characteristic-based decomposition ensemble model for nuclear energy consumption forecasting

We propose a novel mode-characteristic-based decomposition ensemble model for nuclear energy consumption forecasting. Our method is based on the principles of "data-characteristic-based modeling" and "decomposition and ensemble". The model improves on existing decomposition ensemble learning techniques (with "decomposition and ensemble") by using "data-characteristic-based modeling" to forecast the decomposed modes. Ensemble empirical mode decomposition is first used to decompose the original nuclear energy consumption data into a series of comparatively simple modes, reducing the complexity of the data. Then, the extracted modes are thoroughly analyzed to capture hidden data characteristics. These characteristics are used to determine appropriate forecasting models for each mode. Final forecasts are obtained by combining these predicted components using an effective ensemble tool, such as least squares support vector regression. For illustration and verification purposes, we have implemented the proposed model to forecast nuclear energy consumption in China. Our numerical results demonstrate that the novel method significantly outperforms all considered benchmarks. This indicates that it is a very promising tool for forecasting complex and irregular data such as nuclear energy consumption.