Intelligent calibration of wind turbine gearbox dynamic model by integrating LSTM agent model and GA-PSO algorithm

Abstract

A digital twin for a wind turbine gearbox (WTG-DT) is essential for advancing wind farm intelligence and improving the efficiency of wind turbine operation. This study addresses key limitations of existing condition monitoring systems, such as low accuracy and slow parameter updates, by proposing a long short-term memory (LSTM) network to intelligently calibrate model parameters. This ensures the real-time operation and maintenance of wind turbines. A high-fidelity dynamic model is developed and validated by performing frequency analysis of vibration signals collected through the condition monitoring system (CMS), with wind speed and load data from the supervisory control and data acquisition (SCADA) system as inputs. To simplify the complex finite element analysis process, a parameter sensitivity analysis is conducted, and a GA-PSO optimization algorithm, based on genetic algorithms, is applied. These methods generate sufficient training data for constructing an LSTM-based predictive agent model, which is then used to accurately calibrate the virtual model. When applied to 6 MW turbines, this approach significantly improves real-time performance, accuracy, and reliability, enhancing the overall operational efficiency of wind turbines.