Deep Reinforcement Learning and Distributed Solution for Local Network based Optimization of Network Slicing in 5G Communications

Nowadays, the advent of 5G networks has introduced the concept of network slicing to allow the multiple independent networks for coexisting on a shared physical infrastructure. However, the existing Mixed-Integer Linear Programming (MILP) solution suffered inefficient resource utilization which has led to reduced network performance and increased latency. Hence, this research proposes Deep Reinforcement Learning and Distributed Solution for Local Network (DRL-DSLN) based optimization of network slicing in 5G for efficient resource allocation and low latency. The proposed DRL-DSLN framework increases the network performance where it learns from interactions with adapting to changing conditions and makes real time decisions. Initially, the network slicing architecture is introduced to maintain a high performance of network operation by supporting various services. Here, each network slice has a four specific Key Performance Indicators (KPIs) and these are identified to optimize slicing process. Finally, DRL-DSLN are employed to solve the inter-slice allocation and intra-slice allocation problems respectively. From the results, the proposed DRL-DSLN achieved better results in terms of accuracy (98.27%), precision (97.31%), recall (92.94%), and F1-score (97.53%) when compared to existing Harris Hawks Optimization with Convolutional Neural Network (HHO-CNN) model.