B-LSTM ultra-short-term wind power prediction based on LOF data anomaly detection

In today's situation of increasing power load, in order to better promote the utilization of wind energy and further balance power production and market demand, an accurate, efficient and stable prediction model is needed to solve the complex problem of wind power generation. This study introduces a deep learning framework that employs a Bidirectional Long Short-Term Memory (B-LSTM) network. This network discerns the linear associations among various feature variables affecting the output power via the Pearson correlation coefficient. Additionally, it utilizes the Local Outlier Factor (LOF) to identify and filter out data points that are outliers relative to the core dataset. This study presents a dataset compilation from a Shanghai-based wind farm, integrating measurements of wind velocity, temperature, humidity, barometric pressure, and the actual power output from wind turbines at varying tower heights. The model was constructed within the Python execution environment, facilitating ultra-short-term forecasting. Following multiple simulations and debugging iterations, performance was assessed using metrics such as Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). A comparative analysis with alternative electricity forecasting techniques demonstrated the superiority of this approach.