Dyna-Routing: Multi Criteria Reinforcement Learning Routing for Wireless Sensor Networks with Lossy Links

摘要

Finding optimal routes for packet delivery in aWSN is presented as a multi-criteria optimization problem termed Dyna-Routing. Delay in delivering packets to sinks, energy used for transmitting packets, rate of energy depletion in respective nodes and also, properties of the transmission medium are used to form metrics for evaluating routing options. A Dyna Reinforcement Learning algorithm suitable for non deterministic environments is used to speed up learning such that minimal energy is wasted on sub-optimal action choices. Simulation results of Dyna-Routing compared to other machine learning routing approaches show a marked increase in the lifespan of nodes and the network in general.