Tensor Neural Networks Based Optical Fault Localization

We propose a fault localization method based on tensor neural networks to address the challenges of: i) extracting high order interaction features from multidimensional optical monitoring data for fault localization and ii) poor interpretability of neural networks-based models. The method significantly reduces the number of hyperparameters and computational complexity compared to long short-term memory recurrent neural networks and multilayer perceptrons, achieves higher fault localization accuracy, and generates interactive heatmaps that explain the results, providing an efficient and interpretable solution for intelligent operation and maintenance.