Short-term wind forecasting using spatial and temporal wind measurements

Wind intermittency is the main obstacle to wider integration of wind generation. Jay Apt et. al. claim that wind power generation is so unreliable that 100% of scheduled wind generation must be backed up by spinning non-renewable generators. We believe that the requirements for spinning reserve can be reduced by effective short-term forecasting on both spatial and temporal information. Although similar work has been reported [1] in the past, we analyze the available information using different data sets and machine learning methods to provide a comparison of multiple statistical approaches. The training data includes historical wind speed and direction at locations close to the site of interest. The used machine learning methods are sequential minimal optimization (SMO) regression, linear regression, multilayer perceptron, and a Radial Basis Function (RBF) neural network.