Network Slicing Resource Allocation Optimization Based on Multiactor-Attention-Critic Joint With Bidding in Heterogeneous Integrated Network

Abstract

The demand for various types of services is growing rapidly with the development of beyond 5G/6G networks, network slicing (NS) is considered as an effective technology to cope with the multiple services and large traffic demand. In this article, a NS resource allocation optimization algorithm based on multiactor-attention-critic (MAAC) joint with bidding is proposed to guarantee the service satisfaction rate (SSR) while increasing the profit of mobile virtual network operator (MVNO) in heterogeneous integrated networks. First, a pricing and bidding strategies are designed for users with different service requirements and service indexes, and the resource allocation is modeled to maximize the sum of utility of all MVNO subjected to MVNO's pricing and bandwidth constraints accordingly. Second, the optimization problem is analyzed based on the augment Lagrange method, relaxed and has been proved as a convex optimization, and then, the alternating direction multiplier method is adopted to obtain the theoretical upper bound with 32.047 of the network utility. Meanwhile, the gradient descent method with different learning rates is used to accelerate the convergence rate. Third, the MAAC-based algorithm is proposed and the resource allocation procedures are transformed into a partially observable Markov decision process, in which the interactions with multiagent environment are performed accurately. Finally, the simulation results indicate that the network utility of the proposed algorithm can be improved by 25.074% while ensuring the users' SSR. Compared with multiagent deep determination strategy gradient and dueling deep Q network, the network utility by the proposed algorithm can be improved by 6.265% and 39.791%, respectively, up to 27.664, which can be closest to the theoretical upper bound at the greatest extent.