The Internet of Things (IoT) is a platform that connects various electronic systems such as home appliances, vehicles, and medical devices through wired or wireless communications. Without recharging the battery of sensors and mobile systems in IoT networks, their usage time is limited. In order to improve performance with finite battery energy, power management is used to conserve the energy dissipation of sensor networks and mobile systems. This dissertation addresses power management in two categories of systems within IoT: wireless sensor networks (WSNs) and electric vehicles (EVs).

For power management in WSNs, this dissertation develops an algorithm using network coding (NC). When one sender transmits multiple packets to different receivers in a WSN, an NC algorithm reduces transmissions between the sender and the receivers by encoding many packets into one packet. Consequently, the total communication energy between the sender and the receivers is decreased. For further study about real energy gains generated by NC algorithms, we develop a wireless testbed by using mobile devices. Consequently, by varying different network variables such as transmission range of a sender and the number of receivers in the testbed network, we discover network conditions where communication energy saved by NC algorithms is increased. However, NC algorithms spend operational energy overheads for algorithm execution, encoding, and decoding. Hence, our research also shows the threshold conditions where the energy saved by the NC algorithms are larger than the energy overheads with consideration of communication variables or algorithm complexity in order to identify opportunities for energy savings.

For power management of EVs, this dissertation develops an energy-efficient algorithm using neural networks which can be used for power management of EVs' electronic control system. Power management saves energy consumption of the electronic control system by selectively activating electronic control units (ECUs) in the system. However, the energy savings generated by the power management could be less than the energy overheads used for the selective ECU activation and deactivation. Our algorithm experiences events where energy overheads were greater than energy savings and trains neural networks for the experienced events. The neural networks forecast energy-inefficient events and conserve energy overheads based on the predicted events. Our simulation study using real driving datasets shows that the algorithm improves the energy dissipation of the electronic control system by 5% to 7%.