The invention discloses a decentralized adaptive calculation unloading method based on a DDPG algorithm, which is used for a truck fleet system, in the method, the DDPG is composed of an actor network and a critic network, and each part comprises an online main network and a target network; when the method is executed, parameters of all networks are initialized, the DDPG interacts with the environment, and then track tuples generated at all moments serve as training data sets to be stored in an experience playback pool; in the training stage, a small-batch sample containing N track tuples is randomly selected from an experience playback pool and sent to the DDPG to update network parameters; the parameters of the critic main network are updated by adopting a stochastic gradient descent method, and the parameters of the active main network are updated according to the gradient provided by the critic main network. According to the invention, the average energy consumption and the task average calculation time delay of the truck fleet system can be reduced.
本发明公开了一种基于DDPG算法的去中心化自适应计算卸载方法,用于卡车车队系统,该方法中,DDPG由actor网络和critic网络两部分组成,每部分包含一个在线主网和一个目标网络;执行该方法时,初始化所有网络的参数,使DDPG与环境交互,然后将每个时刻生成的轨迹元组作为训练数据集存储到经验回放池中;在训练阶段,从经验回放池中随机选择一个包含N个轨迹元组的小批量样本送到DDPG中以更新网络参数;critic的主网络采用随机梯度下降方法来更新其参数,actor的主网络根据critic主网络提供的梯度来更新其网络的参数。本发明可以降低卡车车队系统的平均能耗和任务平均计算时延。
Decentralized adaptive calculation unloading method based on DDPG algorithm
一种基于DDPG算法的去中心化自适应计算卸载方法
2024-03-01
Patent
Electronic Resource
Chinese
Autopilot Strategy Based on Improved DDPG Algorithm
SAE Technical Papers | 2019
|Fuel cell hybrid power system energy management method based on DDPG algorithm
European Patent Office | 2022
|A Control Method for Quadrotor Based on DDPG
TIBKAT | 2022
|