Insights into the forecasting capability of monthly precipitation via ensemble EEMD-DSE-KELM method

Insights into the forecasting capability of monthly precipitation via ensemble EEMD-DSE-KELM method

Roghayeh Ghasempour1 Hazi Mohammad Azamathulla2 Hassan Sani3

1) 1. Department of Water Resources Engineering, Faculty of Civil Engineering, University of Tabriz
2) 2. Department of Civil and Environmental Engineering, University of the West Indies St. Augustine, Trinidad
3) 1. Department of Water Resources Engineering, Faculty of Civil Engineering, University of Tabriz

IOI Code : XDCB-FEECA
Publication : 7th.International Congress On Civil Engineering, Architecture & Urban Development(7icsau.com)
Abstract :
    Precipitation plays an important role in determining the climate of a region. Accurate precipitation forecasting is an important factor in flood estimation, drought monitoring, irrigation planning, and river basin management. In this study, an integrated forecasting approach based on Ensemble Empirical Mode Decomposition-Differential Symbolic Entropy (EEMD-DSE) and Kernel Extreme Learning Machine (KELM) was applied to enhance the accuracy of monthly precipitation forecasting and reduce the complexity of hybrid forecasting process. In this regard, at first, the original time series was decomposed to several Inherent Mode Functions (IMFs) via EEMD method. Then, DSE was applied to analyze the complexity and reconstruct the subseries. The reconstructed subseries were used as inputs for KELM method. The obtained results proved high capability and efficiency of the applied integrated method in modeling the monthly precipitation. The results showed that time series decomposition based on EEMD-DSE improved models accuracy up to 50% and 15% compared to the single KELM and EEMD-KELM methods. The best evaluation criteria for test series using integrated approach was DC=0.878, R=0.910, and RMSE=0.125. Therefore, it could be deduced that the proposed forecasting model in this study is an effective forecasting model.

Keywords : differential symbolic entropy integrated models kernel extreme learning machine mode components precipitation