Artificial Intelligence Driven Earthquakes Early Detection in Noisy Urban Areas

In this study, we introduce a novel approach for early earthquake detection in urban environments with high ambient noise. By using machine learning techniques to analyze the polarization alterations of light traveling along an existing traffic-carrying optical network, we demonstrate a cost-effective, secure, and efficient solution for detecting primary earthquake wave in noisy conditions. Detecting the primary wave preceding a destructive surface earthquake wave enables the rapid initiation of emergency plans, ensuring timely implementation of earth- quake countermeasures. Our methodology involves collecting large dataset of polarization angular speed evolution along a fiber cable to conduct a Monte Carlo analysis, after integrating the strains induced by car passages over those induced by real earth- quake ground displacement values. This dataset trains a machine learning model that leverages a deep learning architecture based on Long Short - Term Memory layers and attention mechanism. The model’s training and validation show high accuracy rates, implying that additional training is unlikely to yield significant improvements, resulting in a 99% correct detection rate for multi-class classification of all events. The model demonstrates high accuracy in distinguishing between various environmental events, providing accurate early warning signals upon primary wave detection.