Research on edge computing capability optimization of intelligent fusion terminal in distribution area based on deep learning

For the scenario of multiple edge servers and multiple smart meter devices. First, the system model is constructed, and the problem model is to minimize the weighted total cost of delay and energy consumption. Based on DQN, we integrate Actor network and Critic network into it to form a new DDPG algorithm, which adopts deterministic strategy to select actions by approximating the value function model of neural network. According to the actual problem, three elements are set for the deep reinforcement learning algorithm: state, action and reward. Finally, the calculation time of various migration strategies was recorded. The results show that compared with DQN and other algorithms, the proposed DDPG algorithm can obtain a better task migration strategy, and has a significant effect on reducing the total weighted system overhead.