A novel real-time integrated adaptive stable offloading (RIASO) algorithm in multi-access edge computing

Abstract Mobile Edge Computing (MEC) and Mobile Computation Offloading (MCO) help IoT devices with limited computational capabilities and battery by offloading tasks to the nearest resource-rich servers in MEC. Deciding to execute the tasks at the user device or the edge server can be optimized by AI techniques such as Deep Reinforcement Learning (DRL). In this paper, A Real-time Integrated Adaptive Stable Offloading (RIASO) algorithm is proposed based on Lyapunov optimization (LO), Multi Output Learning (MOL), and DRL. In RIASO, a new training policy that regulates the start time of the training procedure can effectively shorten the response time. Also the size of the memory is examined and reduced. RIASO optimizes network data processing capabilities while maintaining long-term data queue stability and average consumed power constraints that makes it adaptable to changes. RIASO achieves high accuracy and efficiency in terms of critical performance metrics such as computation rate, energy consumption, and memory usage.