Background and motivation

Energy efficiency is a fundamental challenge for Internet of Things (IoT) devices operating with limited battery capacity. Existing power-saving mechanisms, such as sleep scheduling and duty cycling, can reduce energy consumption but may also increase communication delays or cause devices to miss important transmissions. Predicting future communication activity could enable more effective energy-saving decisions.

Objective

This thesis will investigate whether a lightweight, pretrained time-series model can predict short-term IoT traffic activity and support energy-efficient device sleep scheduling while satisfying communication-delay requirements.

Proposed work

The work will investigate one or several representative IoT traffic workload(s) and evaluate a lightweight time-series forecasting model, such as IBM TinyTimeMixer. The predicted traffic activity will be used to develop a simple forecasting-based sleep scheduling policy that determines when an IoT device should remain active or enter a low-power state.

The proposed approach will be compared with fixed duty cycling, reactive sleep scheduling, and simple statistical forecasting baselines. Evaluation will use simulation and representative IoT traffic traces. Performance metrics will include total device energy consumption, forecasting accuracy, and the computational and energy overhead introduced by the forecasting model.

Expected outcomes

A simulation framework, an evaluation of the energy–accuracy trade-off, and an assessment of whether lightweight pretrained forecasting provides measurable benefits over simpler sleep scheduling methods.

Contact

Fazal Khan <fazal.khan@kau.se>, Giuseppe Caso <giuseppe.caso@kau.se> or Anna Brunstrom <anna.brunstrom@kau.se>