Adaptive Task Offloading for Mobile Edge Computing With Forecast Information

Mobile Edge Computing (MEC) provides a new approach to satisfy the requirements of delay-sensitive and computation-intensive applications by trading communicational capability for computational capability. Besides seeking for additional physical resources for improving performance, such as Device-to-Device (D2D) paradigm and Intelligent Reflecting Surface (IRS), with Machine Learning (ML) and Artificial Intelligence (AI), information forecast brings a new dimension to unleash the potential of MEC. However, most existing literature does not tailor information forecast algorithm to the MEC context, which diminishes the utility of the forecast information. In this work, we propose a framework for adaptive task offloading based on channel forecast for MEC system. Specifically, we first propose an online resource allocation algorithm with forecast information through Lyapunov optimization. Then, the performance in terms of delay and power consumption is analyzed in closed-form, based on which critical insight towards the requirements on forecast algorithm is obtained. Next, based on Sparse Gaussian Process and Linear Model of Coregionalization (SGP-LMC), we propose a channel forecast algorithm to find a desired trade-off between forecast accuracy and model complexity in terms of maximizing the utility of forecast information. Finally, simulation is performed to verify the performance improvements.