Optimisation of wireless sensor networks using supervision information

Energy saving in wireless sensor networks (WSNs) is a critical problem for diversity of applications. In many scenarios using WSNs, we have incomplete and imperfect prior knowledge about the problem, if this knowledge can be incorporated into the problem-solving process, our performance can be improved. The main goal of this paper is to demonstrate the positive effect of supervision information, i.e., information such as our prior knowledge about the problem domain, on the performance of machine learning in WSNs. To achieve this goal, the routing problem in WSNs is solved with and without supervision information. First, the problem is solved using a very simple routing method, 'Gossiping'. Next, a reinforcement learning-based technique is used to find the most energy-efficient routes. Our methods are analysed theoretically and tested using a simulation. The results are highly promising and show that the utilisation of supervision information can reduce energy consumption by nearly 60%.