公开(公告)号：US20240255584A1

公开(公告)日：2024-08-01

申请号：US18561221

申请日：2023-01-31

Applicant: SOUTHEAST UNIVERSITY

Inventor： Fujin DENG ,  Yufei CHEN ,  Chengkai LIU

IPC: G01R31/54

CPC classification number: G01R31/54

Abstract: The present invention discloses an isolation forest (IF)-based modular multilevel converter (MMC) open-circuit (OC) fault diagnosis method, including the following steps: sampling capacitor voltages of submodules (SM), constructing an isolation tree (IT) according to capacitor voltage data, and calculating a depth of each SM in the IT; and forming an IF based on the ITs, calculating average depths of the SMs in the IF, taking an SM with the minimum average depth as an output of the IT, and finally locating a faulty SM by an output buffer to realize accurate location of the faulty SM. Only capacitor voltages of SMs are involved, and no extra hardware resources are required. System parameters are not involved, construction of a system mathematical model and manual setting of an empirical threshold are not required, and the robustness is high. Faults are located using sparsity and difference of abnormal data based on unsupervised learning, mass data analysis and sample training are not required, and the method has the advantages of linear time complexity, small data volume, simple calculation process, low calculation cost, and the like.