Influence of the input layer signals of ANNs on wind power estimation for a target site: A case study

Abstract

In the scientific literature concerning renewable energies there have been proposals for the use of Artificial Neural Networks (ANNs) as a tool to estimate the mean wind speed at a target station for which only incomplete data series are available. In general terms, the wind speeds recorded at neighbouring reference stations are used as signals of the input layer of multilayer perceptron (MLP) architectures.An analysis is undertaken in this paper of the extent to which estimation of the wind speed, the wind power density and the power output of an installed wind turbine at a target site is affected by the number of reference stations, the degree of correlation between the wind speeds of the reference stations and the target station, the wind direction and the manner in which the direction signal is introduced into the input layer.For the purposes of this study mean hourly wind speeds and directions were used which had been recorded at twenty two weather stations located in the Canary Archipelago (Spain) in 2002 and 2006. The power-wind speed characteristic curves of five wind turbines of different rated power were also used.The MLP architectures were trained using the backpropagation algorithm. Amongst other conclusions reached with the tests performed with the set of data not used in the training nor in the validation of the models is that the estimation errors tend to decrease as the number of reference stations used increases, independently of the existing correlation coefficient values between the reference stations and the candidate station. It was also observed that if input signals of angular wind direction and wind speed are used then the number of reference stations required to achieve a certain decrease in the error is lower than the number required if only wind speed input signals are used. By using multiple reference stations with input signals of angular wind direction and speed, error reductions have been achieved for some stations in the estimation of the wind power density and the power output of wind turbines, with respect to the case of a single reference station and identical input signals, of 75% and 62%, respectively. (C) 2010 Elsevier Ltd. All rights reserved.