Rising energy demands and a growing focus on sustainable development havemade electricity production from wind energy an attractive alternative tofossil fuels. However the natural variability of wind makes it challenging toimplement wind energy into the electrical grid. Accurate and reliable windpower predictions are seen as a key element for an increased penetration ofwind energy.This study presents a set of statistical power prediction models usingthe concept of Markov chains, based on various input parameters, such aswind speed, direction and power output. The models have been trainedand tested using numerical weather predictions and historical data obtainedfrom a meteorological station and wind turbine at Fakken wind farm in thetime period 2. May 2013 - 31. March 2014. Several of the models were foundto have lower NRMSE than the currently used persistent model (19.08 %),with the best performing model having a NRMSE of 16.84 %. This 2.25 %lower NRMSE corresponds to approximately 3 100 000 kWh of the anuallyelectricity production from Fakken wind farm.A statistical analysis of Fakken wind farm showed the majority of windsoccurring from the straits between Arnøya and Lenangsøyra to the southeast and between Reinøya and Lenangsøyra to the south. Winds were alsocommonly seen from southwest and to the northwest, while eastern andnortheastern winds were rarely observed. Westerly winds were found to bemuch more tubulent than other directions, with a generally lower poweroutput observed. This is most likely due to the occurerence of mountainwaves for winds crossing the mountain range to the west.
展开▼
