Fault diagnosis of switching tube in power routers based on IAVOA-PCNN-MSA-SVM

Abstract The switching tube faults in power routers are random and unpredictable. This makes it difficult to extract fault features, resulting in low diagnostic accuracy. To address this issue, a fault diagnosis method for switching tubes in power routers based on IAVOA-PCNN-MSA-SVM is proposed for monitoring and predictive diagnosis. First, the gram-angle field (GAF) encoding method is used. It maps the one-dimensional fault signals of switching tubes in the power routers into two-dimensional fault feature images. These images are inputted into a parallel convolutional neural network (PCNN) with two channels for training. At the end of the PCNN, a multi-headed self-attention mechanism (MSA) and a support vector machine (SVM) classifier are introduced. For the problem of finding the optimal hyperparameter combination in the PCNN-MSA-SVM model, an improved African vultures optimization algorithm (IAVOA) is proposed. It uses multi-strategy fusion for hyperparameter optimization, improving the model’s accuracy in identifying switching tube faults. Simulation results show that this method has a high fault diagnosis accuracy. It can diagnose switching tube faults in power routers with an accuracy of 98.9%. Also, it has a good feature extraction effect. Compared with other methods, it has higher diagnostic accuracy and robustness.