Empirical mode decomposition-based OS-ELM for short-term solar irradiance forecasting: A case study in Hanoi

Abstract

Solar irradiation forecasting is vital for the growth of renewable energy sources. In this paper, we propose a hybrid model that integrates Empirical Mode Decomposition (EMD) and an online sequential extreme learning machine (OS-ELM) for multiple steps ahead forecasting of solar irradiation. Initially, the solar irradiation dataset is processed and cleaned. Then, using the EMD model combined with the autocorrelation function, the cleaned dataset is decomposed into several Intrinsic Mode Functions (IMFs) and white noise, which is removed. Each IMF is subsequently predicted using OS-ELM. The final solar irradiation forecast is derived by aggregating the predictions from all Intrinsic Mode Functions (IMFs). The model's performance was assessed through forecasting solar irradiation in Hanoi, using weather data from 2018. The data was collected at 1-hour intervals and utilized for single-step, 12-step, and 24-step ahead forecasts. The forecasting accuracy of the proposed model was compared with four other models, including both single and hybrid approaches: Bidirectional Long Short-Term Memory network, ELM, OS-ELM, and EMD-ELM. Two evaluation metrics of RMSE and MAE were used to assess the forecasting performance of the models. The computational results show that when the multi-step ahead increases, accuracy decreases. In any case, the proposed method outperforms the others, achieving the lowest error rates at 18,01 W/m² for RMSE and 8,51 W/m² for MAE at 24-step.