An encoder-decoder model with dual-attention mechanism for wind power forecasting

Abstract

Forecasting of intermittent wind power is one of the fundamental challenges for the reliable and controllable operation of systems with wind power. This paper proposes a deep learning model i.e. a dual-attention based encoder-decoder model over Long Short-Term Memory (LSTM) for wind power forecasting. To get the appropriate combination of hyper-parameters, Bayesian optimization technique is implemented on the proposed dual-attention model. To check the accuracy of the proposed model, it has been compared with various state-of-the-art wind power forecasting models. In terms of different error metrics, the proposed model shown better efficiency than the other models. Furthermore, the attention layer's performance against important features have also been analyzed.