NONLINEAR AUTOREGRESSIVE NEURAL NETWORK FOR WIND POWER GENERATION FORECASTING

Abstract

Wind energy is increasingly being utilized globally, in part as it is one of the renewable energy
sources characterized by the lowest cost of electricity production and has experienced a
significant expansion in installed capacity in recent years. Hence forecasting wind behavior, For
example, wind speed is important for energy managers and electricity traders, to overcome the
risk of unpredictability when using wind energy. One of the challenges in integrating wind
power into the electrical grid is its intermittency. One approach to deal with wind intermittency
is forecasting future values of wind power production. But these models rely only on historical
data which is still not convincible enough. Moreover, coping with nonlinear time series data in
forecasting medium to long term with high accuracy is still a challenging task. In this paper,
Nonlinear Autoregressive Neural Network - Long-Short-Term Memory (NARX- LSTM) model
for forecasting wind power has been proposed. This model is able to determine long-term
dependencies in the context of wind power predictions. The proposed model has Root Mean
Square Error of 0.16000 and MAPE of 0.23000. The prediction accuracy of the model is better
than the existing models including BPNN with RMSE of 23.877 and MAPE of 22.032, RBFNN
which has RMSE of 18.729 and MAPE of 16.735, and ARIMA-RBF which has RMSE of 3.00
and MAPE of 2.659. The paper concluded that in terms of correlation error R, the proposed
model obtains a regression value close to 1, which clearly interprets that the model perfectly fits
the data used in both the training, testing, and validation phase.