Abstract: This paper addresses the challenge of capacitor fault monitoring in three-level energy storage inverters by proposing an online monitoring method that integrates signal injection with intelligent algorithms. A low-frequency zero-sequence voltage signal is injected into the closed-loop T-type neutral-point clamped (TNPC) system, introducing controllable voltage fluctuations across the DC bus capacitors and providing a solid data foundation for ripple voltage monitoring. This strategy significantly improves the quality of capacitor voltage data acquisition without affecting the inverter's output line voltage.Based on this, an intelligent identification model is developed using the particle swarm optimization (PSO) algorithm combined with the Backpropagation (BP) neural network. This model effectively integrates the global search capability of PSO and the nonlinear fitting ability of the BP network, overcoming the traditional BP algorithm's tendency to fall into local minima. By preprocessing and extracting features from ripple voltage signals, a high-accuracy capacitor state prediction model is established. Experimental results show that, compared with traditional methods, the proposed approach significantly improves the real-time performance and accuracy of capacitor condition evaluation. The capacitance identification error is within 3%, the equivalent series resistance (ESR) error is within 6%, and the response time is notably reduced. This method provides reliable technical support for online monitoring and maintenance of capacitors in TNPC systems and offers valuable engineering significance for improving the reliability and safety of grid-connected inverter sys-tems.