Deep Reinforcement Learning Based Computing Resource Allocation in Fog Radio Access Networks

The integration of artificial intelligence (AI) with fog radio access networks (F-RANs) has garnered great interest, primarily motivated by the needs for efficient network operation and for ensuring high service availability. Fog access points (F-APs) can help with computation offloading and thereby alleviate the huge computational burdens of terminal devices in F-RANs. However, the overall system energy consumption must to be minimized. As described herein, we propose a computation offloading strategy for industrial internet-of-things (IIoT) devices that is centered around deep reinforcement learning (DRL) based user and F-AP association, which can learn high-dimensional data and which can respond to dynamic changes in the environment. The proposed DRL model adopts a framework that deploys the agent at the user side to address the challenge of high dimensionality in the action space. Specifically, each IIoT device is assigned a dedicated DRL model within the framework, facilitating the identification of an appropriate F-AP based on the environment state. Once the user and F-AP association process is completed, a computationally efficient greedy algorithm is used at each FAP, considering the limited capability, aiding in determining the subset of offloading requests that should be forwarded to the cloud for additional processing. The simulation results showcase the superior performance of the proposed DRL algorithm over traditional algorithms, including the random algorithm and the greedy algorithm, in terms of energy consumption. Under the same operation time, DRL also outperforms the genetic algorithm.