Short-circuit fault diagnosis based on RS-GA-BN for single-phase H-bridge inverters

Short-circuit (SC) of power components in inverters is one of the most serious faults that are vulnerable to occur. It is critical to quickly and accurately detect and locate SC faults in power devices, especially to determine their severity. Therefore, the paper proposes a fault diagnosis algorithm that combined the rough set genetic algorithm (RS-GA) and the Bayesian network (BN). The algorithm uses the RS for attribute reduction and the GA for global optimization to reduce numerous fault attributes and obtains several kinds of reduction results. According to the reduction results, BN diagnostic models are established for optimization, and the optimal RS-GA-BN diagnostic model is determined by an optimization evaluation algorithm based on time and accuracy. More importantly, by the fault attribute values of the optimal reduction attribute of the RS-GA-BN model, the fault severity can be diagnosed, and the fault and its trend can be predicted in advance, which provides an important idea for the prevention of SC. The experimental platform and simulation model are built to obtain fault diagnosis data. The experimental results show that the RS-GA-BN model presents higher fault diagnosis accuracy than the BP neural network, and the experiment verifies the feasibility and correctness of the method.