AI-Driven Dynamic Fault Recognition and Type Prediction System Using Squeeze-Informer for Power Transmission Networks

Over recent times, interconnectivity has increased in modern power transmission networks, requiring intelligent systems for fault detection, which require accurate operational stability and reliable power delivery. Classifying and identifying different fault types under various system conditions is challenging due to the complex, nonlinear, and dynamic nature of fault signals. This publication suggests an intelligent fault analysis and layout in power transmission networks to offer efficiency in fault detection and categorization at the line level in various operating conditions. With the help of the electrical fault detection and classification dataset, the framework provides information about preprocessing techniques, addressing the problem of the missing values with the assistance of multivariate imputation by chained equations (MICE) as well as feature normalization with Rank Gauss Normalization that improves the stability of the model. Hilbert–Huang Transform (HHT) is used to obtain instantaneous frequency and amplitude characteristics of electrical fault signals that characterize nonstationary and nonlinear properties of waveforms. Lastly, SqueezeNet1D with Informer (Squeeze-Informer) model is trained on the features to conduct predictions on the faulty transmission lines — which also involved the prediction of the severity of the types of faults and changing conditions of the network — before and after the fault. The findings derived through the experiment support the claim that the framework offers an accuracy of 99.22%, recall of 99.14%, and precision and F1-score of 99.03, which is why it is an effective tool to use for intelligent monitoring and protection of power transmission infrastructure.