Federated Reinforcement Learning (FRL) Framework for Mobile Data Collectors in IoT Applications

Abstract

Introduction: During Mobile Data Collector (MDC) based data aggregation in Internet of Things (IoT) applications, latency and energy consumption increases, due to incorrect visiting schedules of MDCs. 

Objectives: To determine data collection schedules of MDC, based on IoT sensor’s data generation rate.

Methods: In this paper, a Federated Reinforcement Learning (FRL) framework is proposed for MDCs in IoT applications. In this technique, the data generation rate of IoT nodes are learned by applying FRL framework from which the nodes are classified as Emergency, Normal, Less Frequent and Rare. Then depending on the category of the nodes, the visiting schedule and stopping time of MDCs are determined. 

Results: The proposed FRM framework is implemented in NS2 and it has been shown the proposed MDC-FRL framework reduces the data collection latency and energy consumption and improves the accuracy.