Impact of vibration on wind turbine efficiency and LSTM-based power conversion prediction

During the long-term operation of wind power generation systems, the impact of mechanical vibration on energy conversion efficiency is often overlooked. Existing studies mostly use vibration signals as a means of fault warning, lacking a systematic analysis of the quantitative relationship between vibration characteristics and power generation efficiency. This study, based on Supervisory Control and Data Acquisition (SCADA) data and high-frequency vibration monitoring from a wind farm, extracts vibration features from both time domain (e.g., root mean square, peak value, skewness, and kurtosis) and frequency domain (e.g., dominant frequency and spectral energy ratio). Through Pearson and Spearman correlation analyses, as well as a comparative time series analysis of high-vibration intervals (revealing an average efficiency drop of 3.5%), it is demonstrated that intensified vibrations significantly reduce generation efficiency and increase output fluctuations. Furthermore, a dual-layer LSTM prediction model is proposed, integrating wind speed, wind direction, temperature, and vibration features. The training process is optimized using a sliding window strategy, Dropout regularization, and early stopping. On the test set, the model achieves an RMSE of 0.035 and a MAPE of 3.6%, outperforming support vector machines (SVM), random forests, and single-layer GRU models by 20%–40% in accuracy. Finally, an integrated “monitoring–prediction–warning–control” framework is proposed to support real-time deployment and intelligent operation and maintenance (O&M), offering a practical solution for wind farm health management and O&M optimization.