AN AFFORDABLE ALTERNATIVE TO EARTHQUAKE EARLY WARNING SYSTEMS USING RECENT ADVANCES IN MACHINE LEARNING AND MICROELECTRONICS

In this work, I propose the Affordable Earthquake Early Warning (AEEW) system, a cost-effective alternative to traditional Earthquake Early Warning (EEW) systems. AEEW utilizes low-cost accelerometer arrays and is regulated by several neural network committees, some of which update locally in real-time via Field Programmable Gate Arrays (FPGAs). For obvious reasons, the implementation of effective EEW systems has long been a goal of Scientific and Engineering communities. Currently, several EEW systems are in use, but the costly nature of ultra-precise instrumentation required creates an insurmountable entry barrier for many governments. Compounding this issue, multidisciplinary research shows that economically developing nations are substantially more socioeconomically vulnerable to natural disasters, and often report fatality rates an order of magnitude higher in comparison to wealthy countries. Therefore, the lack of an EEW system that is both affordable and robust likely exacerbates relative vulnerability for at-risk populations. Machine Learning (ML) is a powerful tool for event prediction that has been widely applied to the problem of earthquake detection with much success. Algorithms recently developed for EEW systems, which are adaptable to AEEW systems, were compared to current ML literature and possible improvements were identified. Additionally, recent advancements in microelectronics allow for precise (< 1 mm/s^2) measurement of seismic acceleration by relatively inexpensive components known as Micro Electro Mechanical System (MEMS) accelerometers. These components have previously been suggested and used as the basis for an EEW system, though environmental noise, communication delay, cost, and inconsistent reliability all limit functionality in current implementations. I suggest an Artificial Neural Network (ANN) committee receiving inputs from sparse distributions of precision-variant, low-power sensors in geographically optimized locations to solve noise and reliability related issues. Additionally, self-regulatory neural networks running locally on FPGAs resolve communication delay issues and further increase reliability. Finally, projections for AEEW systems were compared to their EEW counterparts on the basis of cost and reliability. The primary focus of this research is to provide access to reliable and affordable alternatives to EEW systems. Results from this research could be used as a reference for the future development of affordable disaster detection warning systems.