A fault diagnosis and location method for power grid simulators based on voltage threshold and MCNN

To accommodate the testing requirements of high-power wind turbines, this paper designs a power grid simulator topology and investigates fault diagnosis and localization methods by integrating mathematical models and neural networks. To address the drawback of lengthy computation times associated with intelligent diagnostic methods, this paper employs a threshold-based approach using voltage mathematical models to achieve rapid preliminary diagnostics. To address the positioning challenges brought about by symmetrical structures, a multi-layer convolutional neural network (MCNN) model is utilized to achieve accurate positioning. To tackle the issue of insufficient fault samples, a sliding window technique and frequency domain transformation methods are applied to expand the sample set, enabling the diagnosis and localization of 36 types of faults. This paper builds an inverter-side model of the power grid simulator using Simulink to verify the proposed method. And the diagnostic accuracy rate reaches 100%, and the overall localization accuracy exceeds 96%.