Design and application of a wind power prediction system

The main problems encountered in wind power predictions are using no multi-tupled inputs, considering different time and error scales, utilizing no different data mining techniques, including no multiple data presentation techniques and no improvement ratios in presenting the prediction results and using no datasets containing seasonal effects. In this study, a wind power prediction system is designed and developed for the purpose of overcoming the existing problems partially and/or fully in wind power predictions. In addition to, it is also applied in different fields to test the performances. The wind power prediction system designed included two models called agglomerative hierarchical clustering and k-nearest neighbor classification. The similarity analyses of 81 cities in Turkey are made in terms of monthly average wind speed by the first model and the cities suitable for electricity generation based on wind power are determined. In the study, wind speed and power are predicted at 10-min time intervals using meteorological input data in n-tupled input space by the second model. Apart from these, the first and the second models developed are used effectively in similarity analysis of monthly average insolation period data and prediction of yaw position, pitch angle and solar radiation, respectively. The results achieved show that the models developed gave successful results in terms of prediction accuracy, independency for application and problem, capability of analyzing any dataset in the field of renewable energy sources. It is also expected that the software platform developed can be successfully used for educational purposes providing students to select different clustering and classification methods with different data sets, input tuples and time intervals in their exercises.