An EMD Based Polynomial Kernel Methodology for superior Wind Power Prediction.

Abstract

This paper proposes low complexity Empirical mode decomposition trained by Kernel based (KEMD) algorithm for wind power prediction for various time horizon such as ten minutes to five hours interval for California wind farm. For a comparative performance analysis, another two forecasting model named as Empirical mode decomposition with pseudo inverse neural network and Pseudo Inverse neural network with Legendre functions and RBF units, which is further optimized by Firefly algorithm (FFA) is described here. Kernel based pseudo inverse algorithm is proposed because it eliminates the involvement of the hidden layers in each iteration, which helps in return to reduce the computational complexity and generates more precise response in prediction purpose. In the other two models the weights which are used between the hidden layer and the output neuron are obtained by PINN which is also known as Moore-Penrose pseudo inverse algorithm. This proposed KEMD trained by kernel based pseudo inverse algorithm has a very good and precise prediction of wind power. This model has been proved by doing several observations for various seasons which has been demonstrated in the results and simulation section.