A novel Antarctic wind turbine power prediction method based on TTAO–CNN–Bi-LSTM–SAM that considers meteorological factors under extreme conditions

The environmental pollution in Antarctica, which is caused mainly by fuel-based energy consumption, should not be underestimated. To protect the Antarctic environment, clean energy technologies must be urgently explored. By utilizing the abundant wind energy resources in Antarctica, accurate output power prediction for Antarctic Wind Turbines (AWTs) becomes essential. To address the challenges associated with Antarctic wind turbines power prediction, this study first proposes a novel data preprocessing method—Thompson–Improved based spatial clustering of applications with noise (DBSCAN)–Quartile Interpolation (TIDQ), which is developed based on the physical operational characteristics of Antarctic wind turbines, for correcting anomalies and missing values in real meteorological and power data. Subsequently, key meteorological factors with significant influence on wind power generation are identified through correlation coefficient analysis and incorporated into the input features of the prediction model. Finally, a new Antarctic wind turbine power prediction framework is constructed by integrating the Triangulation Topology Aggregation Optimizer (TTAO), Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (Bi-LSTM), and Self-Attention Mechanism (SAM). The TTAO algorithm is employed to optimize neural network hyperparameters and mitigate overfitting. Empirical results demonstrate that the proposed model significantly enhances prediction accuracy an reduces computational time compared with other commonly used optimization algorithms. Results show that the proposed TTAO–CNN–Bi-LSTM–SAM model achieves the lowest RMSE and MAE among all methods, reducing RMSE by up to 6.94 and MAE by up to 6.22 compared with other cases. The computational time remains within an acceptable range, demonstrating both higher prediction accuracy and faster training efficiency.