Wind Speed Forecasting Using Improved Random Vector Functional Link Network

Abstract

This paper proposes an improved Random Vector Functional Link Network (RVFL) for better performance in regression problems. The model applies the Chebyshev expansion to transform the direct links of the RVFL providing better mapping of nonlinear functions when compared with RVFL model. Two wind speed datasets are used for performance comparison with other models. The application of Chebyshev expansion in the RVFL enables the RVFL to have a lower number of activation nodes reducing its size with better performance. The models are also tested with their ensembled version by applying the empirical mode decomposition (EMD). © 2018 IEEE.