Resumen
There is a growing demand for environmentally friendly energy. With an increase in the level of implementation of this type of power plants in the energy sector, the need for forecasting the generation of these plants is growing. Electricity generation forecasting is a necessary process for any power plant, its complexity varies depending on the type of plant. One of the most technically difficult is forecasting the generation of wind farms. The complexity of this process can be associated with the strict dependence of the station on weather conditions and how the windmills will shade each other under certain weather conditions. Also, the complexity of these calculations lies in the fact that meteorological towers are not placed directly on wind turbines, so it is necessary to calculate a mathematical model for forecasting meteorological data based on data from nearby meteorological towers. To predict the data, neural networks are used with the method of inverse error distribution to update the weights of the perceptron. This article discusses an approach to forecasting the generation of wind farms using a neural network. Mathematical models suitable for the process are considered and their accuracy is compared. The architecture of neurons, the type of neural network suitable for such calculations, the method for calculating weights in it, the activation functions of the network, and the method for calculating the gradient are described.