Adversarial learning‐based multi‐timescale network resource management in multi‐mode green IoT network for smart building

Abstract

Multi-mode green internet of things (IoT) network that integrates multiple communication media can well meet data transmission and processing demands of low-carbon smart building. However, network resource management optimisation including joint optimisation of gateway and channel selection still faces technical challenges such as differentiated quality of service (QoS) demand guarantee, coupling between optimisation problems with different timescales, and adversary caused by multi-device competition. To address these challenges, an adversarial learning-based multi-timescale network resource management algorithm for multi-mode green IoT is proposed. Specifically, the minimisation problem of weighted difference between energy consumption and throughput under the long-term queuing delay constraints is formulated to achieve differentiated QoS guarantee. Large-timescale gateway selection is decoupled from small-timescale channel selection by establishing matching preferences based on empirical performance, and optimised by using bilateral matching with quota. Finally, an exponential-weight algorithm for exploration and exploitation (EXP3)-based small-timescale channel selection algorithm is proposed to achieve adversary awareness. Simulation results demonstrate that compared with asynchronous greedy matching algorithm and auction-based many-to-many matching algorithm, the proposed algorithm performs superior in terms of energy consumption and throughput.