Reinforcement Learning based Optimized Multi-path Load Balancing for QoS Provisioning in IoT

Abstract

In an IoT system, to achieve optimization, performance should be an equal concern along with satisfying the growing requirements and demand of solutions, for real time, especially time critical applications. Some of these applications are complex to accommodate current solutions that is concerned with multivariate and multiobjective performance optimizations. Hence smart learning of the system helps identify the nuances in the system that affects the performance of the system. The main goal of the protocols used in the network layer is to perform routing process and forwarding packets by recognizing and achieving best decisions to optimize network performance to achieve better Quality of Service (QoS) for the application. Prolonging the lifetime of the network keeps the network on its purpose active and achieves QoS. Hence in this paper we have proposed an algorithm for load balancing in an uncertain IoT network by choosing multi-path for data transmissions. We categorize the data into various classes that can use various levels of optimized paths. Using the reinforcement learning algorithm – Q-learning approach and the QoS parameters as the hyper parameters, the algorithm we have proposed is compared with the conventional Q-routing algorithm and proved the improvements of the proposed algorithm in network longevity and throughput.