An optimized task offloading strategy based on deep reinforcement learning combined with channel reliability prediction

Mobile edge computing provides a new solution to meet the computing demands of emerging applications such as the industrial Internet, which cannot be fully met by the resources on the device side. However, the edge nodes and transmission channels are not always completely reliable. When the network channel is unreliable, it is easy to cause task transmission failure and lower service quality. In this paper, an optimized task offloading strategy based on deep reinforcement learning combined with channel reliability prediction is proposed, named deep deterministic policy gradient-based strategy combined with hindsight experience replay and LSTM (HL-DDPG). The HL-DDPG strategy uses long short-term memory (LSTM) to mine the time dependence between the channel reliability states. The task offloading is then modeled using the Markov decision process (MDP), and the joint optimization problem is solved using the DRL method based on the Actor-Critic framework. Meanwhile, Hindsight experience replay (HER) is used to improve the learning ability of the algorithm. The experimental results show that compared with four baseline algorithms, HL-DDPG has a lower overall offloading error probability and a lower task timeout rate, which effectively improves the reliability of the edge offloading system and reduces the risk of task transmission failure.