Calculation unloading and resource allocation method based on deep reinforcement learning

Abstract

The invention discloses a calculation unloading and resource allocation method based on deep reinforcement learning. The problem of global cost minimization is processed by using a deep reinforcement learning method, namely, a double-deep Q learning method, and the method can obtain an optimal calculation unloading and resource allocation strategy in a time-varying channel state and a random task arrival environment; and the deep neural network is used as an optimizer of a value function, so that dimension disasters caused by a high-dimensional state space can be reduced, and the convergence speed can be increased.

Description

technical field [0001] The invention belongs to the field of wireless network communication, and in particular relates to a calculation unloading and resource allocation method based on deep reinforcement learning. Background technique [0002] With the development of wireless network technology, especially the development of 5G and the emergence of 6G, in wireless network communication, how to meet higher quality of service (QoS) for communication and computing becomes more and more important. Although the computing power of wireless devices (WD) has made great progress with the improvement of its processors and manufacturing process standards, its processing power is still insufficient to meet QoS when faced with a large number of computing-intensive or delay-sensitive computing tasks. . As the computing data generated by WD grows exponentially with strict deadline constraints, WD's battery capacity and resource constraints remain bottlenecks. Therefore, how to meet the ...

AI Technical Summary

Problems solved by technology

When the task is large or divisible, the binary offloading scheme greatly reduces the unloading efficiency. Regarding the resource allocation scheme, most technologies consider the allocation of computing resources on the MEC server side, but few applications consider the WD and MEC servers. Bandwidth resource allocation between, and most applications on resource allocation assume that the channel state is fixed

But in actual application scenarios, the channel state between WD and MEC server is time-varying, and the tasks generated by WD are not fixed

Images