Wind Power Forecasting Based on Bidirectional LSTM Integrated with Chaos Enhancement Theory

Wind energy prediction is crucial for efficient management and dispatch of power systems, and accurate prediction can optimize the integration of renewable energy sources and can be a good solution for energy security. This study proposes a chaos-enhanced bi-directional LSTM model for wind power forecasting that incorporates chaotic signals from Lorentz attractors to enhance the feature representation of time series data and combines the powerful contextual information capturing capability of bi-directional LSTM. The goal of this research is to develop a model that outperforms traditional as well as current mainstream machine learning aspects of prediction methods. The method splices the generated slices of chaotic signals with the input sequences in the feature dimension and captures the forward and backward time dependencies using bi-directional LSTM, and finally feeds its output into the fully connected layer for final prediction. The experimental results show that the proposed model outperforms several current mainstream machine prediction models with a MAPE of only 0.5979% and an RMSE of 4.2, which demonstrates the effectiveness and superiority of the model in wind power prediction.