CNN–LSTM–AM: A power prediction model for offshore wind turbines

This study introduces a power forecasting model, the convolutional neural network (CNN)–long short-term memory (LSTM)–attention mechanism (AM) algorithm (CNN–LSTM–AM), designed to predict the power of offshore wind turbines based on data collected by a SCADA system. The model employs a timestep parameterisation approach for offshore wind turbine prediction, facilitating automatic partitioning of the training dataset and simplifying the training process. A CNN–LSTM–AM network was presented to predict the power of offshore wind turbines using signals from multiple sensors. A variable–control comparison was conducted to complete the sensitivity analysis of the sensors, which determined the most suitable sensor group for power prediction. The model achieved a maximum improvement of 13.77% in power prediction compared to existing deep learning algorithms. The results indicate that the hub and rear-end temperatures of the high-speed shaft of the gearboxes are crucial for offshore wind power prediction. Overall, the findings of this study contribute to the operation and maintenance of offshore wind turbines and the management of offshore wind farms.