FAULT DIAGNOSE METHOD OF OPEN-CIRCUIT OF PV INVERTERS BASED ON IMPROVED GAF-SE-RESNET
Aiming at the problem that the one-dimensional time series signals of PV inverters cannot adequately capture the temporal and local features when they are input into the convolutional neural network,a PV inverter open-circuit fault diagnosis model based on the combination of Gramain angular fields(GAF)and improved deep residual network(ResNet)is proposed.Utilizing GAF encoding method with two channels,the one-dimensional current signal is mapped into a two-dimensional fault feature image with distinct pixel distributions.Using the feature images as the input to ResNet preserves the temporal correlation of the data.ResNet incorporates residual modules in convolutional neural networks to mitigate overfitting.An improved version of the residual module includes compression and squeeze-and-excitation(SE)attention mechanisms for image compression and feature reuse,enhancing important feature information.These enhancements enable ResNet to delve deeper into image information and fully capture local features.Combining the Swish function and Ranger optimizer to optimize ResNet,the training difficulty of the model is significantly reduced.The experimental results show that the method has an accuracy of 99.41%for diagnosing open circuit faults in PV inverters,and has better feature extraction effect and diagnosis speed compared with other models.
万方数据
维普
