Abstract
Wind energy is one of the fastest-growing renewable sources both globally and in Brazil. A key challenge in this sector is the accurate estimation of energy production, which is affected by wind variability and other operational factors. The power curve, which relates wind speed to electrical power, is a fundamental tool for evaluating wind turbine performance. This study proposes an artificial neural network (ANN) to predict power curve using operational SCADA data from a wind farm in southern Brazil. A robust pre-processing strategy was employed, featuring outlier filtering using the 3-sigma rule and air density normalization of wind speed as per the IEC 61400-12-1 standard. The proposed ANN model consists of four hidden layers and an output layer, all using the hyperbolic tangent activation function. Using both air density-normalized wind speed and turbulence intensity as inputs, the model was trained with the Adam optimizer to capture their complex non-linear relationship with power output. The model’s performance was benchmarked against the traditional IEC 61400-12-1 binning method, with the ANN demonstrating high predictive capability and achieving coefficients of determination (R2) greater than 0.96. These results underscore the potential of data-driven ANN models as a robust alternative to standard methods for monitoring and optimizing wind energy production.
Funding Information
  • Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES), Brazil, under the CAPES–IEEE Open Access agreement

This publication has 32 references indexed in Scilit: