A Machine Learning-Driven Smart Optical Network Grid for Earthquake Early Warning

Optical fiber networks, commonly recognized for their role in data transmission, have the potential to be extended beyond their conventional use. These networks could serve as wide distributed arrays of sensors for earthquakes early detection by monitoring and identifying specific evolution patterns of the light's state of polarization (SOP) caused by the strain induced by external perturbations on fiber cables. We propose a centralized smart grid system that take advantage of the existing terrestrial network infrastructure, offering an efficient solution for earthquake early warnings and early emergency responses by detecting earthquake primary waves (P-waves). Our focus is on monitoring changes in light's polarization due to the strain induced by primary waves' arrivals, and subsequently use this data to refine and evaluate a machine learning model to interpret and detect these changes. This paper presents a novel neural network model based on Temporal Convolutional Network employed on our smart grid sensing approach. Tested on real earthquake data, our method achieves accurate detection of primary waves with 98% of accuracy rate.