Improving Short-Term Wind Power Prediction with Neural Network and ICA Algorithm and Input FeatureSelection

According to this fact that wind is now a part of global energy portfolio and dueto unreliable and discontinuous production of wind energy; prediction of windpower value is proposed as a main necessity. In recent years, various methods havebeen proposed for wind power prediction. In this paper the prediction structureinvolves feature selection and use of Artificial Neural Network (ANN). In this paper,feature selection tool is applied in filtering of inappropriate and irrelevant inputs ofneural network and is performed on the biases of mutual information. Afterdetermining appropriate inputs, the wind power value for the next 24-hours ispredicted using neural network in which BP algorithm and PSO and ICAevolutionary algorithms are used as training algorithm. With investigation andcompare numerical results, better performance of PSO and ICA evolutionaryalgorithm is deduced with respect to BP algorithm. More accurate survey will resultin more proper efficiency of imperialist competitive algorithm (ICA) in comparisonto swarm particle algorithm. Thus, in this paper; accuracy of the wind powerprediction for the next 24-hours is improved considerably using mutual informationand providing an irrelevancy filter for reducing the input dimension by eliminatingthe irrelevant candidates and more effectively using Imperialist competitiveevolutionary algorithm for training the neural network.

Improving Short-Term Wind Power Prediction with Neural Network and ICA Algorithm and Input FeatureSelection Keywords:

Neural Networks , Wind Power Prediction , PSO Algorithm , ICA Algorithm , Feature Selection , Mutual Information