Publication

Abstract : IoT systems play a vital role in the realtime monitoring of environmental hazards for disaster management applications. One such scenario we see in this paper is the I0T-based landslide monitoring application. The sensor data from the IoT nodes in the deployment site is sent to the landslide data management center (DMC) located 230 miles away from the site. Landslides are susceptible during monsoon seasons, and it is during this time the IoT network connections to the DMC become unstable due to environmental and climatic conditions. Hence the IoT nodes should have a failover network and at the same time they should have edge intelligence to choose the network depending on dynamic QoS. The chosen solution should also apply to the Green IoT nodes that has less hardware resource and function using ambient energy like solar. This paper proposes a solution that uses edge intelligence to choose among multiple networks in the IoT nodes using using multi-arm bandits. We formulate our problem into approximate, contextual, and adversarial bandits that does edge-based online learning in the IoT nodes. We simulate these algorithms with the actual dataset collected from the field and show their effectiveness compared to the naive baseline algorithm. The results show the approximate bandits and contextual bandits perform better than other algorithms. The proposed solution applies to ultra-low memory devices with as little as 8KB of main memory. This solution might help many outdoor applications like volcano monitoring, flood, and forest fire monitoring to have network failover during rough environmental conditions. © 2023 IEEE.