Analyzing the energy efficient path in Wireless Sensor Network using Machine Learning

Abstract

As the sensor nodes are energy constrained, an important factor for successful implementation of a Wireless Sensor Network (WSN) is designing energy efficient routing protocols and improving its lifetime. Network life time has been described in many ways such as   the time when the network lost its connectivity or the time when the first node gets disconnected. Whatever may be the description, the main focus of many researchers is to design algorithms that enable the network to perform continuously for a longer duration. So, improving the energy efficiency and increasing the network lifetime are the two key issues in WSN routing. Because of the intelligent nature and learning capacity, reinforcement learning (RL) algorithms are very suitable for complex distributed problems such as routing in WSN. RL is a subclass of Machine Learning techniques.  It can be used to choose the best forwarding node for transmitting data in multipath routing protocols. A survey has been made in this paper regarding the implementation of RL techniques to solve routing problems in WSN. Also, an algorithm has been proposed which is a modified version of original Directed Diffusion (DD) protocol. The proposed algorithm uses Q-learning technique which is a special class of RL. Also, the significance of balancing the exploration and exploitation rate during path finding in Q-learning has been demonstrated using an experiment implemented in python. The result of the experiment shows that if exploration-exploitation rate is properly balanced, it always yields an optimum value of the reward and thus path found from source to the destination is efficient.