Controlador del ángulo de guiñada de un aerogenerador que utiliza redes neuronales artificiales implementadas en un sistema integrado

Abstract

In this research, artificial neural networks (ANN) developed in python are compared and later compiled in a Raspberry pi 4 to generate a predictive wind direction signal as wind turbine control system input, to maximize the capture of wind power. A set of 12 weather station measured variables are used to feed the neural network, including time, PM10, PM25, and Ozone as secondary variables that will allow enriching the predictive factors of the neural network, the variables NO, NO2, NOX, and SO2, as auxiliary variables that will allow strengthening the validation of the behavior of the network and finally the variables Wind Speed, temperature, relative humidity and wind direction as main variables that will increase the prediction efficiency and with this, to complete partial dependence between the variables is analyzed to improve the ANN convergence time on the embedded system, as future work, it will allow the testing of a control system including control actuators to optimize the network