● Development of the CAOO-CNN-GRU model for fault diagnosis in hydro-turbine runner; ● Introducing the chaotic sequence to enhance the AOO algorithm; ● Optimizing CNN-GRU hyperparameters based on CAOO; ● Researching hydro-turbine runner failures from an acoustic signal perspective. The wide-load operation of hydro-turbines in new power systems introduces complexity to the blade flow state, and runner failure is a significant contributor to hydro-turbine failure. In this paper, from the perspective of acoustics, a hydro-turbine runner fault identification model based on the Chaotic Animated Oat Optimization algorithm (CAOO), which combines the convolutional neural network (CNN) and the gated recurrent unit (GRU), is proposed. The features of acoustic signals related to runner faults are adaptively extracted and downscaled using a CNN. GRU enhances the time series modeling capability. Chaotic sequences improve the initial population of AOO to accelerate convergence and enhance the glossary. Subsequently, the hyperparameters of the CNN-GRU model are optimized by CAOO. A hydro-turbine failure experimental bench is built to train and validate the model, and the experimental results show that the training and testing accuracy of the CAOO-CNN-GRU model reach 99.5% and 96.1%, respectively. Compared with AOO-CNN-GRU, the accuracy is improved by 11.4% and 5%, respectively, showing better stability and generalization ability. This study can serve as a valuable supplement to existing hydro-turbine condition monitoring and fault diagnosis systems.
Lisheng et al. (Sun,) studied this question.