haku: @supervisor Janhunen, Tomi / yhteensä: 234
viite: 2 / 234
| Tekijä: | Pandey, Binda |
| Työn nimi: | Adaptive Learning For Mobile Network Management |
| Julkaisutyyppi: | Diplomityö |
| Julkaisuvuosi: | 2016 |
| Sivut: | vii + 54 Kieli: eng |
| Koulu/Laitos/Osasto: | Perustieteiden korkeakoulu |
| Oppiaine: | Computer Science (SCI3042) |
| Valvoja: | Janhunen, Tomi |
| Ohjaaja: | Rintanen, Jussi |
| Elektroninen julkaisu: | http://urn.fi/URN:NBN:fi:aalto-201612226199 |
| Sijainti: | P1 Ark Aalto 6063 | Arkisto |
| Avainsanat: | mobile networks network management reinforcement learning intelligent control |
| Tiivistelmä (eng): | There is an increasing demand for better mobile and network services in current 4G and upcoming 5G networks. As a result, mobile network operators are finding difficulties to meet the challenging requirements and are compelled to explore better ways for enhancing the network capacity, improving the coverage, as well as lowering their expenditures. 5G networks can have a very high number of base stations, and it costs time and money to configure all of them optimally by human operators. Therefore, the current network operations management practice will not be able to handle the network in the future. Hence, there will be a need for automation in order to make the network adaptive to the changing environment. In this thesis, we present a method based on Reinforcement Learning for automating parts of the management of mobile networks. We define a learning algorithm, based on Q-learning that controls the parameters of base stations according to the changing environment and maximizes the quality of service for mobile devices. The learning algorithm chooses the best policy for providing optimal coverage and capacity in the network. We discuss different methods for taking actions, scoring them, and abstracting the networks' state. The learning algorithm is tested against a simulated LTE network and the results are compared against a base-line model, which is a fixed-TXP model with no control. We do the experiments based on above-mentioned strategies and compare them in order to evaluate their impact on learning. |
| ED: | 2017-01-08 |
INSSI tietueen numero: 55238
+ lisää koriin
INSSI