Deep Reinforcement Learning Based Partial Offloading of Multi-DAG Applications in the Internet of Vehicles

A partial offloading algorithm based on deep reinforcement learning is proposed for the problem of multiple Directed Acyclic Graph (DAG) type applications in the edge computing for Internet of Vehicles (IoV) environment. Firstly, a partial offloading model for DAG applications is proposed with the goal of maximizing the comprehensive utility of delay and energy consumption. Then, an execution priority algorithm is used to convert DAG applications into a sequential structure for determining the execution priority of each sub-module in DAG applications. On this basis, a seq2seq network based on RNN is designed as policy network. Finally, the partial offloading problem of multiple DAG applications is converted into a single DAG application, and the partial offloading of multiple DAG type applications is implemented based on deep reinforcement learning. Experimental results show that the proposed algorithm achieves better comprehensive utility than baseline algorithms in the same offloading scenario.