Comparing the accuracy of regression and artificial intelligence methods in estimating daily speed of wind in the Sistan region

Abstract

This paper aims at comparing the accuracy of regression methods, artificial intelligence methods, and phase-neurotic interpretation method in estimating wind speed in the Sistan region. To this end, we used the daily weather information obtained from Zabol synoptic stations during a five-year period (2010-2015). MATLAB software was used for modeling based on artificial neural network. On the other hand, DATA FIT software was used for modeling based on regression methods. Methods’ accuracies were estimated using error square mean statistics, comparison indexes, and error mean. Based on sensitivity analysis results; variables such as daily temperature mean, mean relative humidity, sunshine hours, and evaporation from pool were regarded as input variables of regression and artificial intelligence methods. Wind speed was considered as output variable. Based on the results, mean daily temperature and mean relative humidity had the most and the least effect on wind speed in Sistan (0.42 and 0.25 respectively). Neurophasic method with Gaussian function was the most accurate method in estimating wind speed (error squares mean of 2.56). The same statistic for regression method is 4.44. The correlation of regression method (0.45 and 0.51) is less than those of multilayer Perceptron method and Neuro-phasic method (0.51 and 0.52). So, it is suggested that Neurophasic method be used for more accurate estimating wind speed in Sistan region. With accurate estimation of this variable, we can hinder the devastative effects of wind and use it as an effective source of energy.

Keywords

• sensitivity analysis
• error squares mean
• neuro-phasic method
• detection coefficient