Resource Allocation for Vertical Industries in a Smart City Using Deep Reinforcement Learning
Master thesis
Permanent lenke
https://hdl.handle.net/11250/3087982Utgivelsesdato
2023Metadata
Vis full innførselSamlinger
- Studentoppgaver (TN-IDE) [823]
Sammendrag
This Master’s thesis addresses the problem of resource allocation in Network FunctionVirtualization (NFV) within the context of a smart city considering isolation. Theobjective of the thesis is to minimize the long-term cost and maximize the revenue andacceptance rate of the infrastructure provider. A resource allocation model is proposed,utilizing a state-of-the-art proximal policy optimization (PPO) training algorithm, andits performance is evaluated under decreasing available resources.In this thesis, we first introduce the smart city and the basic NFV and reinforcementlearning (RL) concepts. Then we review the existing literature on resource allocation inNFV. Further, we introduce our problem definition and solution as well as our proposedNFV-compatible isolation levels.The results we produce demonstrate that the proposed solution exhibits increasedperformance overall compared to a leading heuristic algorithm called global resourcecapacity (GRC), especially in terms of the long-term revenue-to-cost ratio (LTRCR) ofat the most 20%, and in terms of the 25x reduction in time spent allocating per slicerequest (SR).