부산대학교 도서관

In order to locate the open circuit fault of the power switching device of the neutral point clamped three-level inverter more efficiently and accurately, a BP neural network fault diagnosis algorithm optimized based on Improved Grey Wolf optimizer (IGWO) and adaptive moment estimation (ADAM) is proposed. Wavelet packet four layer decomposition is used to decompose the fault current signal, which can get more fine detail components; The improved gray wolf algorithm can help BP neural network determine the initial value. Finally, the adaptive moment estimation algorithm is used to dynamically update the learning rate, which can make BP neural network match the learning rate and gradient value in the gradient search process. The learning rate is small in the place with large search gradient and large in the place with small gradient search. The simulation results show that compared with traditional BP neural network and IGWO-BPNN, IGWO-ADAM-BPNN diagnosis model has the advantages of fast convergence speed, high classification accuracy and strong generalization ability. It is suitable for NPC three-level inverter fault diagnosis.