COLLABORATIVE COMPUTING UNLOADING OF BLOCKCHAIN INTERNET OF THINGS BASED ON MULTI-AGENT DRL

Due to the slow convergence speed, poor robustness and unstable performance of deep reinforcement learning algorithm, a collaborative computing unloading algorithm based on multi-agent deep reinforce learning framework for blockchain IoT is proposed. An efficient multi-agent deep reinforcement learning algorithm was designed, and an initialization method based on agent strategy was proposed, which avoided useless exploration in the initial stage of agent training and greatly reduced the time required to achieve stable performance in agent training. The coalition learning mechanism was introduced, and a decentralized network was constructed for the agent to improve the adaptability of the algorithm to the dynamic environment. The simulation results show that the algorithm can effectively improve the robustness and convergence speed.