Wavelet-Enhanced Sequence-To Modeling With Attention Mechanism for Short-Term Wind Power Forecasting

Abstract

Electricity load forecasting is crucial to managing electric systems, especially loads produced from renewable energy sources since the load from renewable energy sources varies when compared with nonrenewable sources. Turkey is producing an increasing amount of electricity from wind energy every day. The aim of this study is to introduce a hybrid deep learning model based on sequence-to-sequence learning (seq-2-seq), attention mechanisms, and wavelet transformation. Long Short-Term Memory (LSTM), Gated Recurrent Unit, and Bidirectional Long Short-Term Memory (BiLSTM) are used as decoders and encoders in the seq-2-seq model. We proposed six different models. All models are univariate type, requiring only the data itself. The model can be used on any wind farms without requiring the meteorological data. We test the proposed model on four different wind farms in Turkey: Soma, Biga, Balikesir, and Mersin. We utilize four different performance metrics to test the model's performance: mean squared error (MSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and coefficient of determinations (R2). The best model is seen as Wavelet-Seq2Seq-BiLSTM-LSTM at Biga Wind Farm, which achieved the best performance with a MAE of 0.127, an MSE of 0.001, a MAPE of 0.28, and an R2 of 0.997.