Short-term Wind Energy Forecasting Using Attention-based Encoder Decoder GRU Framework

With growing penetration of renewable energy in power grids, the significance of accurate forecasting for renewable energy such as wind power is increasing. However, wind power exhibits stochastic temporal variations, which makes the forecasting task as a challenge. To alleviate this problem, this paper proposes an attention-based encoder decoder neural network. The overcome the long sequence dependencies vanishing problem in existing recurrent neural network, the attention mechanism is introduced to dynamically model the relations among features in global context. The variant of long-short term memory (LSTM) neural network, gated recurrent unit, is introduced to act as the encoder and decoder. To improve model's robustness and stability in long sequence forecasting, teacher forcing mechanism is adopted in training process. Further, multiple features including time-period, temperature and wind speed are delicately processed. A real-world wind farm power generation dataset is used to verify the performance of proposed model. Compared with other benchmark models. The numerical results demonstrate that the proposed model outperforms other benchmarks and enables to accurately forecast wind power generation in different time scale.