Fault prediction method of hydro turbine generator based on a new long short term memory strategy

To address the needs for turbine unit failure prediction (TUFP), we propose a TUFP-Kalman-LSTM deep learning network that integrates the Kalman filter for preprocessing with a Long Short Term Memory (LSTM) neural network. To optimize the Hyper-parameters, we introduced the Quantum Satin Bowerbird Optimizer (QSBO) algorithm, establishing a QSBO-based method for Hyper-parameter tuning of the TUFP-Kalman-LSTM network. This novel TUFP-Kalman-LSTM-QSBO method was applied to the online monitoring data of unit #2 at a hydropower station in Hubei province. The results show that this approach effectively predicted vibration trends, enabling intelligent prediction of turbine generator units.