An Approach to Integrate Reinforcement Learning in Wireless Sensor Network to Evade Congestion

Machine learning is the most trending technology heeded nowadays due to its applications and problem-solving mechanisms which made humans extend its application to sectors like cloud computing, wireless sensors and many more. WSN is versatile technology owing to its aspects like miniature in size, cost effective and easy to deploy but facing issues like energy deficiency, security and congestion control. This paper addresses a critical issue in WSN congestion control using the RL (Q learning) algorithm. Also describes machine learning approaches to relevant congestion control issues and parameters that have been improved. And suggested CARRL Protocol to obtain optimum path and update Q-table with rewarded routes on time basis, and compared QL and CARRL performance with end-to-end delay (EED) and packet delivery ratio (PDR), as well as discussed, the algorithm's throughput and learning rate. Furthermore, how the discount factor helps to obtain the above factors is explained, and simulations are performed in the network simulator environment with parameters mentioned. Finally, the CARRL Protocol approach performance is compared with shortest path algorithm and Q-learning algorithm with factors mentioned above along with number of node and results are depicted in simulations section and noted preferable performance improvements in delivering the packets as compared to Q-learning but still more experiments need to be done with huge number of nodes with packets