Prediction of ultra-short-term wind power based on BBO-KELM method

Abstract

For ultrashort-term wind power prediction, an optimized extreme learning machine method based on biogeography-based optimization (BBO-KELM) is proposed. The kernel extreme learning machine (KELM) method only uses the kernel function to represent the unknown nonlinear feature map of the hidden layer and does not need to select the number of nodes of the hidden layer. Meanwhile, the output weight of the network is calculated by the regularized least squares algorithm. The BBO algorithm, which is a new evolutionary algorithm (EA) motivated by biogeography, which is the study of the distribution of biological species through time and space, is efficient in solving high dimensional, multiobjective optimization problems. In this paper, the KELM method is optimized using the BBO algorithm to optimize the selection of input variable sets, the parameters of the kernel function, and the Tikhonov regularization coefficient, so as to further improve the learning performance of the KELM method. To verify the effectiveness of the BBO-KELM method proposed in this paper, the BBO-KELM method is applied to ultrashort-term wind power prediction research in different regions and is compared with benchmark methods such as persistence, neural networks, support vector machine, extreme learning machine (ELM), and other optimized ELM (O-ELM) or KELM (O-KELM) methods such as BBO-ELM, particle swarm optimization (PSO)-ELM, differential evolution-KELM, simulated annealing-KELM, and PSO-KELM, under the same conditions. Experimental results show that the BBO-KELM methods with cosine migration can give better prediction accuracy; in addition, in the proposed method, the parameters of the kernel function do not need to be selected by trial-and-error and the relevant input variables can be automatically selected, improving the generalization capability.