Short-term Wind Power Probabilistic Forecasting Based on Attention Mechanism and Bidirectional LSTM

In order to quantify the uncertainty of regional wind power generation, this paper proposes a probabilistic forecasting method of wind power based on attention mechanism and bidirectional long short-term memory (Bi-LSTM). First, data cleaning is performed on the historical wind power data, and a feature set for prediction model training is constructed. Then, to make the prediction model selectively focus on the information with strong correlation, while ignoring the information with weak correlation, an attention mechanism is introduced to capture the dependencies and relationships between different parts of the input. Afterwards, a Bi-LSTM is used to construct the forecasting model, which realized the long-term memory of the power sequence effectively reduces the occurrence of falling into local optimum. Finally, the currently generated deterministic forecasts are compared with past forecasts, the most similar past forecasts are selected, and the corresponding observations are used to generate probabilistic forecasts.