Dynamic power management strategies for a sensor node optimised by reinforcement learning

Dynamic power management DPM is considered in wireless sensor networks for improving the energy efficiency of sensor nodes. DPM usually includes two classical problems, dynamic operating mode OM management and adaptive transmission mechanism. In this paper, we propose a new model of Markov decision process that combines dynamic OM management and adaptive transmission mechanism. In addition, a fragment transmission scheme is further integrated to reduce the probability of retransmission failure and improve the data transmission rate. The model takes into account the performance criteria being the expected cost of synthesising the per-packet energy consumption, the buffer overflow, the fragment cost and the energy consumption of operating mode switching. Reinforcement learning algorithm is subsequently proposed to search for the optimal strategies. Furthermore, a state-clustering approach is given to increase the learning speed and lessen the storage requirements. Finally, an example is presented to illustrate the effectiveness of the proposed method and show that the energy consumption is well-balanced dynamically under the optimised policy, while the throughput decreases only slightly. Therefore, the lifetime of the node can be extended under constrained resources.