Adaptive Energy Management Scheme in Real-Time Energy Harvesting Embedded Systems

The main purpose of this thesis was to develop an online adaptive energy management scheme for energy harvesting embedded systems. Two energy prediction schemes were used, namely Exponentially Weighted Moving Average (EWMA) and Adaptive Forward Prediction (AFP), to schedule all the tasks with least deadline miss rate. The AFP scheme has a mean relative error of 6-10% which is much lower than exponentially weighted moving average (EWMA) algorithm with an error of 30%. The large difference in the error percentage between the two prediction algorithms is due to the adaptive nature of AFP as it tracks small changes in input signal and dynamically adjusts itself to the changes incurring smaller error percentage. On the other hand, EWMA algorithm requires prior knowledge of the signal from the previous day which doesn’t remain constant, thus introducing large prediction error. The proposed algorithm is executed in two parts- Firstly, an offline energy management algorithm using EWMA was developed which decides the speed at which the task should be executed depending on the energy availability. Secondly, using AFP algorithm the tasks speed and start time was dynamically adjusted according to the difference in the energy predicted by both the prediction algorithms during runtime. The results show that by using the proposed adaptive technique the deadline miss rate of the tasks was decreased by 15-30% in addition to the results accomplished by initial scheduling depending on the extra/less amount of energy predicted by AFP