考虑逆变器动态行为累积的改进高斯混合模型故障诊断方法OA

The open-circuit fault signals of grid-connected inverters exhibit non-Gaussian and multi-modal distribution characteristics,and they are also susceptible to noise and variations in operational conditions,resulting in inadequate performance of the traditional diagnostic methods.To solve these problems,a robust fault diagnosis method with en-hanced feature discriminability is proposed in this paper.First,discrete state equations are constructed based on a uni-fied circuit model under grid-connected inverter faults to derive the system state space representations for fault feature extraction.Second,kernel mean embedding is employed to measure the distribution discrepancies between fault fea-tures,thus realizing an effective evaluation on the feature discriminability.On this basis,a Bayesian information criteri-on(BIC)-optimized Gaussian mixture model(GMM)is studied to characterize the nonlinear mapping relationships be-tween fault features and categories.In addition,the optimal hyperparameters in GMM are selected according to the BIC,so as to enhance the model's fault classification performance.Experimental results demonstrate that the proposed meth-od achieves excellent performance in terms of accuracy and robustness,with the highest accuracy reaching 99.92%.