Forecast-driven task offloading for reliable and adaptive mobile edge computing

doi:10.1016/j.jnca.2026.104494

Forecast-driven task offloading for reliable and adaptive mobile edge computing

2026-04-21

0

PRE

AI

DOI：10.1016/j.jnca.2026.104494

Original

Original request for help

Save

Abstract

En 中文

Mobile edge computing (MEC) is critical for latency-sensitive applications operating under highly dynamic workloads and reliability fluctuations. Existing optimization-based methods depend on static models, while learning-based methods remain reactive and lack foresight into future load or failure conditions. Although recent prediction-based methods introduce limited predictive signals into offloading decisions, their simple or task-specific predictors and complex decision architectures prevent them from capturing multi-scale temporal patterns or supporting real-time responsiveness. This paper proposes a forecast-driven task offloading framework that closes the loop between short-horizon forecasted modeling and online decision optimization. The framework employs TimesNet as a multi-step time-series learner to extract multi-period temporal structures from server load and failure dynamics, producing richer and more reliable forward-looking context than conventional pointwise forecasting. These forecasted results are then incorporated into an online offloading policy based on LinUCB (Linear Upper Confidence Bound), enabling lightweight, context-aware, and rapidly adaptive task offloading with sublinear regret while fully exploiting forecasted future conditions. By explicitly coupling forecasting with online learning, the proposed design enables proactive and reliability-aware offloading under nonstationary MEC conditions. Extensive experiments on real-world datasets show that our method reduces task failures and system cost while achieving faster convergence than state-of-the-art baselines. These results confirm the advantages of coupling multi-period forecasting with lightweight online decision-making for reliable and adaptive MEC systems.

Keywords:

forecast-driven task offloading

mobile edge computing

multi-period forecasting

online decision optimization

reliability-aware offloading

Forecast-driven task offloading for reliable and adaptive mobile edge computing

Abstract

Journal

Organization