Dynamic Routing Algorithm with Q-learning for Internet of things with Delayed Estimator

Abstract

With the popularity of the Internet of things (IoT), tremendous objects are connected to the network, making the network topology complex. Mechanical engineering achieves intelligent identification, positioning, tracking, monitoring and management of engineering machinery through the IoT, where information exchange and communication require novel intelligent routing algorithms as traditional routing algorithms are unfit for current network environment. Q-routing implemented a dynamic adjustment which was based on the network environment by combining the Q-learning algorithm. However, Q-routing is a highly random network environment and leads to a decline in performance because of overestimation of values. To solve the problem, we propose an algorithm called Delayed Q-routing (DQ-routing), which uses two sets of value functions to carry out random delayed updates so as to reduce the overestimation of the value function and improve the rate of convergence. The experiments indicate that DQ-routing algorithm gets well performance in several problems.