A novel hybrid methodology for wind speed and solar irradiance forecasting based on improved whale optimized regularized extreme learning machine

Abstract

With rising demand for electricity, integrating renewable energy sources into power networks has become a key challenge. The fast incorporation of clean energy sources, particularly solar and wind power, into the existing power grid in the last several years has raised a major problem in controlling and managing the power grid due to the intermittent nature of these sources. Therefore, in order to ensure the safe RES integration providing high-quality power at a fair price and for the secure and reliable functioning of electrical systems, a precise one-day-ahead solar irradiation and wind speed forecast is essential for a stable and safe hybrid energy system. Here, we propose a novel hybrid methodology for wind speed and solar irradiance forecasting. The proposed integrated model employs complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) to decompose time series data into a sequence of intrinsic mode functions of lower complexity. Further, permutation entropy is employed to extract the complexity of IMFs for filtering and reconstruction of decomposed components to alleviate the difficulty of direct modeling. Then, a unique swarm intelligence technique, the non-linear dimension learning Hunting Whale Optimization Algorithm (NDLHWOA), is devised to optimize regularized extreme learning machine model parameters to capture the implicit information of each reconstructed sub-series. By integrating a non-linear convergence parameter and the dimension learning hunting approach, the performance of WOA can be drastically enhanced, leading to premature convergence, enhanced population variety, and effective global search. The final prediction outcome is obtained by summing the individual reconstructed sub-series prediction outcomes. To evaluate its efficacy, the proposed model is compared to five well-established models. The evaluation criteria demonstrate that the suggested method outperforms the existing methods in terms of prediction accuracy and stability, thus confirming that a hybrid forecasting model approach combining an efficient decomposition method with a simplified but efficient parameter-optimized neural network can enhance its accuracy and stability.

Keywords: Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN); Dimension Learning Hunting (DLH); Permutation Entropy (PE); Regularized Extreme Learning Machines (RELM); Solar Irradiance forecasting; Whale Optimization algorithm; Wind Speed forecasting.

Algorithm 1

Empirical Mode Decomposition (EMD) Algorithm

Fig. 1

Variation of parameter ’a’ value in conventional WOA and with exponential decay approach presented in this study.

Fig. 2

Flow diagram of NDLHWOA.

Fig. 3

Comparative analysis of the convergence curves of the suggested algorithms with other algorithms over various mathematical functions.

Fig. 4

Proposed CEEMDAN-PE-NDLHWOA-RELM framework.

Fig. 5

a) Site location b) Weather station used in this study.

Fig. 6

Wind speed data for the entire predicted period.

Fig. 7

Solar irradiance data for the entire predicted period.

Fig. 8

Reconstruction error with EMD, EEMD, and CEEMDAN algorithms for wind speed and solar irradiance decomposition of Dataset-2.

Fig. 9

CEEMDAN decomposition of wind speed and solar irradiance series for all datasets.

Fig. 10

Recombination results for the CEEMDAN-PE deconstructed datasets.

Fig. 11

Wind speed prediction curves derived by the various models for all datasets.

Fig. 12

Comparison of different models based on RMSE and MAE for wind speed forecasting.

Fig. 13

Comparison of models based on for wind speed forecasting.

Fig. 14

Scatter plot of the proposed model in wind speed forecasting for all dataset.

Fig. 15

Solar Irradiance prediction curves derived by the various models for all Datasets.

Fig. 16

Comparison of solar irradiance forecasting models based on RMSE and MAE.

Fig. 17

Comparison of models based on for solar irradiance predictions.

Fig. 18

Scatter plot of the proposed model in solar irradiance forecasting for all datasets.