Neural networks based prediction of wind gusts Vedrana Spudi´ c, Matej Mari ´ c, Nedjeljko Peri´ c University of Zagreb Faculty of Electrical Engineering and Computing Unska 3, 10000 Zagreb, Croatia E-mail: vedrana.spudic@fer.hr, matej.maric@itgcentar.hr, nedjeljko.peric@fer.hr Abstract With increase of wind power penetration into electrical grid its influence on grid conditions grows. Therefore, new demands upon wind power plants emerge. Such demands are track- ing of the imposed power set point and respect- ing imposed constraints on rate of change of the injected power. Due to large blade inertia conventional wind turbine pitch control system is not always capable to meet this objective. This is especially pronounced during large wind gusts. The basis for enhancement of ability of the control system to meet new objectives could be based on feed forward information about up- coming wind gusts. In this paper the strategy of spatially distributed wind predictions based on neural networks is proposed and validated on measurements obtained at test site. The mea- surements from an upwind meteorological mast were used to predict the upcoming wind gust on downwind position. 1. Introduction Wind generated electricity is a highly unpre- dictable intermittent power source. Tradition- ally, main objective in wind energy production was maximizing the power output. Since the wind speed changes in time and wind power is proportional to cube value of wind speed expe- rienced by wind turbine rotor, maximization of power output inevitably leads to power fluctua- tions. With increase of wind power penetration into electrical grids its influence on grid conditions increases and demands upon produced power characteristics emerge. According to devel- opment of national grid codes it is expected that demands upon wind farms will grow fur- ther and that tracking of the imposed power set point and power rate of change will be re- quired. Classical wind turbine control system which relies on wind turbine speed and power measurements is not entirely appropriate for such demands. Main reason for it is slow pitch actuator response in comparison to dynamics of sudden change of wind speed during wind gusts. Large wind gusts cause output power to exceed its set-point due to slow reaction of pitching mechanism caused by large blade in- ertia. This introduces an overshoot in set-point tracking which could be avoided by employing better control technique. Disturbance on set- point tracking is specially pronounced in situa- tions when wind gust causes wind speed to in- creases from a value beneath to a value above nominal wind speed. In such cases the dead time, which is caused by releasing the brakes of pitching mechanisms, further deteriorates the system response. Also, additional objective in wind turbine control is alleviation of loads ex- perienced by wind turbines. During wind gusts wind turbines experience extreme loads. De- scribed problems motivate the employment of control algorithm which would treat wind gusts as a separate operating scenario with different control strategy than what might be called nor- mal operation. To identify this operating sce- nario and to enable the adequate use of slow actuators wind gust predictions are required. The aim of this research is to provide a sim- ple way of predicting abrupt changes in wind speed which lead to large changes in electrical power injected into the grid. If the information about upcoming wind gust would be given, wind turbine control system could react in advance by adapting its objective from normal operation (maximizing energy output) to alleviation of dis- turbance on power output (or some load reduc- tion strategy). In this paper a methodology for spatially dis- tributed predictions of wind gusts is proposed. 2