Spain enjoys excellent wind resources. In practically all regions, a multitude of iniciatives have arisen to install a great number of wind parks. The integration of a significant amount of wind power will cause a serious impact on the efficiency, operation and control of the Spanish power system, which must be analized using sophisticated tools. In the Load Frequency Control, high wind power penetration expected into system generation mix may compromise the efficiency of operation of overall system, causing frequency variations and, even, the system might be lead to collapse. These variations must be efficiently compensated by generator regulators to avoid large frequency excursions. For the succesfull control of frequency it is necessary to have accurate short-term forecast of wind power generation a few seconds to a few minutes ahead. Forecasts based on a recurrent high-order neural network are introduced in a regulator model of synchronous generator with the objetive to reduce frequency variations in a power system.