Gaussian-Weighted Trend-Seasonal Decomposition Interactive Attention LSTM for Wind Turbine Drivetrain State Forecasting

Accurate prediction of the operational states of wind turbines is crucial for improving the stability of their operation and the utilization efficiency of wind power. Currently, most data-driven studies focus primarily on forecasting the output power of wind turbines, without addressing other important operational data such as rotor loads. Therefore, this paper proposes the Gaussian-Weighted Trend-Seasonal Decomposition Interactive Attention LSTM model for predicting the operational states of wind turbines. Given the diverse and complex nature of wind farm data, the Gaussian-Weighted Trend-Seasonal Decom-position is designed to decompose the features of different time patterns within the wind farm data, with Gaussian weighting applied to emphasize information from adjacent time points. Additionally, the Interactive Attention LSTM is designed to separately predict the trend and seasonal components, using the attention mechanism to enable interaction between the current states and historical time points, thus enhancing the contribution of historical information to future feature predictions. Experimental results demonstrate that the proposed model outperforms other models in terms of prediction accuracy for multiple wind turbine states.