Selection of Input Variables for the Prediction of Wind Speed in Wind Farms Based on Genetic Algorithm

Abstract

Wind power generation is increasing very rapidly all around the world. The available wind energy depends on the wind speed, which is a random variable and depends on the location and weather conditions. For the wind-farm operator, this uncertainty creates difficulty in the system scheduling and energy dispatching. Due to this there is a need for predicting the wind speed, i.e., power in advance. The first consideration in predictive wind speed is the selection of appropriate input variables. Many variables are available for wind speed prediction, and some of them cannot be omitted without a significant loss of information. The collection of field data, on the other hand, is both time-consuming and expensive. This becomes more complex and costly as the number of variable's increases. Due to this, rigorous methods are needed to detect which variables are essential and important and, which are not. Appropriate selection of input variables is not only important for modeling objectives as such, but also to ensure reliable decision-support in unit commitment and energy policy-making. In this paper, application of genetic algorithms is explored to automatically select the relevant input variables for Artificial Neural Networks (ANNs). The applied database consisted of 12 different variables, which are measured from Jaipur (Rajasthan). The measured variables are a combination of temperature, wind and atmospheric. The genetic algorithm selects different combinations of input variables and applies them to back-propagation ANN network for wind speed prediction. The ANN predicts the wind speed using the selected variable set and then the various combinations are compared based on their predictive power. With this technique, the number of input variables could be reduced from 13 to 8. In addition, the prediction success increases by the maximum 5 percent. The data requirement, calculation time, memory requirement, complexity of the ANN, cost of measuring equipments, etc. reduces due to removal of irrelevant data from the process.