Vibration Fault Diagnosis for Hydroelectric Generating Units Based on the Modified Cuckoo Search Algorithm and Evidence Theory

Abstract

Background: In view of the complex system structure and uncertain factors in the fault diagnosis of hydroelectric generating units (HGU), it is a difficult problem to design the diagnosis method rationally. Objective: An attempt is made to employ multi-source feature information to improve the accuracy of fault diagnosis, and the effectiveness of the proposed scheme is verified by using a diagnostic example. Methods: Through the research on recent papers and patents related to fault diagnosis of the HGU, a hybrid scheme based on the modified cuckoo search algorithm, back-propagation (BP) neural network and evidence theory are proposed. For this modified version named cuckoo search with fitness information (CSF), the step factor is adaptively tuned using the fitness value. Next, three diagnostic models based on BP neural network trained by CSF are used for primary diagnosis. These diagnostic results are then used as the independent evidence, and the fusion decision is made by using evidence theory. Results: Experimental results show that CSF algorithm is better than the original cuckoo search (CS) and its three variants, and the hybrid method has the highest diagnostic accuracy. Conclusion: The proposed hybrid scheme has strong robustness and fault tolerance, and can effectively classify the vibration faults of hydroelectric generating units