Abstract: Voltage source static synchronous compensator (VSC-STATCOM) can improve power quality by reducing peak and valley voltage fluctuations, effectively controlling harmonics and compensating reactive power. Among them, the normal operation of VSC-STATCOM is a prerequisite for the safe and reliable operation of the inverter. The diagnosis of IGBT open-circuit faults of the inverter under the grid-connected situation needs to consider the high complexity of the fault signal due to the dynamic change of the load, the interaction of the various components of the complex system, as well as the influence of high background noise, which puts forward a higher requirement for the extraction and identification of the fault signal features. A combined fault diagnosis method integrating the bimodal decomposition signal processing method and the residual attention network RES-SA model is proposed to address this problem. The method utilizes three-phase compensated current signals from various open-circuit faults in a grid-connected inverter VSC-STATCOM as fault samples. To address the limitations of single-modal multi-scale signal decomposition for extracting complex signal features amidst high background noise, it employs two decomposition methods (CEEMDAN and SSD). This approach enables two-modal multi-scale decomposition of the same fault signals and integrates the three-phase current bi-modal multi-scale components into the input feature matrix. Such integration supports the RES-SA deep learning model in extracting hidden features. The simulation results of different schemes show that the proposed method has strong feature extraction capability and good noise immunity, and has a high recognition rate of open-circuit faults of grid-connected inverters IGBT.