Air-space-ground integrated network intelligent switching method based on DQN

Abstract

The invention discloses an air-space-ground integrated network intelligent switching method based on a DQN, and belongs to the technical field of mobility management. The method comprises the following steps: constructing a space-ground integrated scene based on a satellite and a ground station, and modeling five state factors when a space-ground integrated network is switched; receiving signal strength RSS, available bit rate ABR, relative speed, transmission delay Delay and network overhead; then, inputting the modeling vectors of the five state factors into DQN for experience learning, and reversely updating Q network parameters by means of mean square errors through action selection and experience playback; and finally, correcting and estimating a current real-time state measured by a user through Kalman filtering, inputting the current real-time state into the Q network with updated parameters, outputting a Q value, and carrying out switching decision by utilizing a greedy method, namely selecting a base station corresponding to the maximum action from the obtained Q value for connection. According to the invention, the switching rate and the switching signal overhead are reduced, and the return effectiveness is improved.

Description

technical field [0001] The invention belongs to the technical field of network mobility management, in particular to a DQN (Deep Q-Network)-based intelligent switching method for air-space-ground integrated network. Background technique [0002] In the current ground-based system, it is difficult for the 5G network to achieve effective coverage in areas such as oceans, polar regions, high mountains, and frequent natural disasters; therefore, space-based and space-based systems with wider coverage and more flexible layout are obtained as a supplement to the ground-based system. In the future, air-space-ground integrated integrated networking will become an important trend. [0003] In the air-space-ground integrated network, the three networks can fully reflect their respective advantages, get rid of the shackles of environmental terrain and other factors, and realize uninterrupted services around the clock. However, problems such as unbalanced resource allocation and unstab...

AI Technical Summary

Problems solved by technology

[0004] In the existing research, the most important handover trigger condition is the received signal strength RSS, and the handover methods around RSS include: the handover method based on the maximum RSS, the handover method based on the RSS single threshold (RSS-T) or the RSS hysteresis threshold. Handover method (RSS-H), etc. Although the above methods can effectively reduce the ping-pong effect of handover, it is difficult for a single reference factor to adapt to all communication scenarios, so multi-attribute handover decisions that consider various network attributes emerge as the times require; but many The attribute switching decision is only for the current moment, and the problem is generally NP-hard, which is difficult to solve

Classical Q-learning has been applied to switching strategies in research, but because Q-learning can only deal with discrete variable problems and problems with small state sets; for continuous variable problems, although it can be quantified, the quantified value itself has errors. At the same time, if the state set is too large, Q-learning will maintain a Q table with a huge dimension, which will cost a lot

Images