Offshore wind Power Output Prediction Method Based on Bi-LSTM and Time Attention Mechanism

As an important source of renewable energy, offshore wind power has been widely concerned and developed rapidly in recent years. Given that offshore wind power shows strong randomness and volatility, accurate prediction of offshore wind power output is of crucial importance for promoting the effective integration of renewable energy and the existing power grid. Aiming at the problem of unstable prediction of offshore wind power output, this paper proposes a hybrid framework based on Bi-LSTM (Bi-directional Long Short-Term Memory) and time attention mechanism. Firstly, it is necessary to normalize the historical data of offshore wind power. The Bi-directional Long Short-Term Memory (Bi-LSTM) model is used to capture the correlation between the data before and after, and the complex dynamic characteristics in the time series data of offshore wind power are modeled and analyzed. Secondly, the wind power output data is decomposed by STL (Seasonal-Trend Decomposition using Loess) to obtain the wind power output trend, and the trend part data is modeled by Bi-LSTM. Then, the MLP (Multilayer Perceptron) model is used to extract the external factors (such as weather, market fluctuations, etc.) that affect the offshore wind power output; Finally, the processed offshore wind power output, wind power output trend and external influencing factor data are fused to obtain the prediction results. Compared with other prediction methods, the prediction results of the method in this paper have at least a reduction of 1.2460 in root mean square error (RMSE), at least a reduction of 0.6002 in mean absolute error (MAE), and at least an increase of 0.0134 in the coefficient of determination R2.