A Method of Controlling Routing Action Based on Multi-agent Reinforcement Learning Routing Policy

Abstract

The present invention relates to the field of information technology, and discloses a method for controlling routing actions based on multi-agent reinforcement learning routing strategies, including: training reinforcement learning models, the reinforcement learning models update the decision values ​​of routing nodes using Q-learning algorithms, and combine The updated decision value uses a policy gradient algorithm to update policy parameters; according to the target node where the routing node forwards the data packet and the network load in the communication network where the routing node is located, the policy parameter is determined using the reinforcement learning model; according to the The policy parameter is used to determine the outgoing link of the routing node. The present invention aims at dynamically changing network connection modes and network loads, and routing nodes can adjust routing strategies in time, select the appropriate shortest path according to the target node of the data packet, and finally greatly reduce the average delivery time of the data packet.

Description

technical field [0001] The invention relates to the field of information technology, in particular to a method for controlling routing actions based on multi-agent reinforcement learning routing strategies. Background technique [0002] Packet routing in communication networks is an important application problem in sequential decision making. A communication network consists of a set of nodes and the links connecting these nodes. Data center networks and the Internet can be seen as real-world examples of communication networks. In a communication network, information is passed between nodes in the form of data packets. Routing is the decision-making process that guides how data packets pass through a series of intermediate nodes from the initial node to the target node. Usually, there are multiple paths for a data packet to choose from in the communication network, and the choice of the path usually determines the average delivery time of the data packet. [0003] At pres...

AI Technical Summary

Problems solved by technology

These simulated network models ignore many important network characteristics, such as dynamically changing network loads and mobile users, so the routing choices made under these models often cannot minimize the average delivery time of data packets

Images