Resource Allocation Method for Wireless Internet of Things Based on Probabilistic Transfer Deep Reinforcement Learning

Abstract

The invention discloses a wireless Internet of Things resource allocation method based on probabilistic transfer deep reinforcement learning, which distributes decision-making agents in each edge server, so that each agent only needs to make decisions for the users it serves That is, the decision variable space is greatly reduced, and the decision delay is also reduced. At the same time, a service migration model based on a distributed partially observable Markov decision process is proposed, which overcomes the problem that each agent can observe The state information of the state is limited, so that the decision-making cannot reach the optimal solution.

Description

technical field [0001] The invention relates to the technical field of networks and Internet of Things, in particular to a method for allocating wireless Internet of Things resources based on probabilistic transfer deep reinforcement learning. Background technique [0002] MEC (Multi-access Edge Computing) is an edge cloud platform that provides a new network architecture by combining with the operator's network (the data plane function is the point of integration), and uses the wireless access network to provide nearby IT services required by telecom users and cloud computing functions, thereby creating a carrier-class service environment with high performance, low latency and high bandwidth, so that consumers can enjoy high-quality business experience. [0003] Such as figure 1 As shown, the MEC network of an application scenario consists of N car users driving in a fixed direction M base stations and the edge server to which each base station belongs composition. ...

AI Technical Summary

Problems solved by technology

[0005] (1) The decision variable space will increase linearly with the increase of the number of users, making it difficult for the algorithm to converge

[0006] (2) The increase of decision variables will also increase the decision-making delay sharply, which cannot guarantee the real-time performance of the system

[0007] (3) Global information is required for each decision, so the system must have a central node to collect the status information of all devices in real time, which will undoubtedly increase the transmission burden of the network and the status information search delay

But the ensuing problem is that the state information that each agent can observe is limited, so that the decision-making cannot reach the optimal solution.

Images