Page 103 - ITU Journal Future and evolving technologies Volume 2 (2021), Issue 4 – AI and machine learning solutions in 5G and future networks

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 4

network structures through a network topology
Proportion of the links with Balanced Loads
optimization model, this system enhances the
60% utilization of network resources and reduces
48.70%
50% 43.00% operators' investments. The advent of the 5G era
40% 35.50% brings the world booming network traffic and more
30% complicated network structures, and AI-based
18.10%
20% network topology optimization is superior to other
10% methods for its outstanding capability and
0% practicability in dealing with complex networks. It
Initial Status Basic Topology Topology ensures sustainable network development and
Optimization Restructuring Restructuring
Based on Links Based on Links guarantees the return on investment ratio for
With 500 Meters With 1000 operators. The AI-based network topology
Meters
optimization system introduced in this paper is
Fig. 8 – Proportion of links with balanced load in different proposed based on ITU AI standards, and features:
optimization stages
1) Creativity: It addresses difficulties and
The above-mentioned solution introduces a promotes the development of intelligent
complete network topology optimization system for networks by applying cutting-edge AI
effective resolution of network topology problems, technologies to operator networks.
as shown in Fig. 9 below.
2) Enhanced traffic forecast algorithm: It enables
more accurate traffic forecasting.
3) A complete network topology analysis and
optimization structure: The concepts of
neighbor, node removing method, and three-
step network topology optimization that are
introduced for the first to the industry
effectively accelerate the network topology
analysis. Moreover, the network restructuring
Fig. 9 – A complete network topology optimization system
fixes the defects of existing network topologies,
4. CONCLUSION paving the way for future network topology
optimization.
Traffic forecasting and network load balancing are
always under the spotlight of operators. During the The challenge organized by ITU has offered a great
study, all tests prove that the forecast accuracy and opportunity for building a cross-field ecosystem,
efficiency of the optimized LSTM are higher than and operators and ITU should continue to make
those of ARIMA, LightGBM, Prophet, and DeepAR. joint efforts (e.g. encourage crowd-funding AI
Therefore, it can provide better support for algorithms) to resolve common problems. For
operators' resource investment. The network example, building a middleware platform to open
topology optimization model proposed in this paper data and solve problems together, organizing more
is optimized based on the actual size of the network competitions and building the ecosystem for
that we worked on in the study. In view of different developers for closer collaboration.
locations, periods of time, network sizes, and
network characteristics, the model needs to be REFERENCES
optimized in accordance with different topology [1] Lampiris E., Zhang J., Simeone O., et al.
reshaping rules tailored based on the nodes and Fundamental Limits of Wireless Caching
routing of the network to be optimized. This model under Uneven-Capacity Channels[J]. 2019.
is vital to network topology optimization, as it can
enhance the resource utilization by over 30%. [2] He Z., Jian H. Application of Multilink
Aggregation and Load Balancing in Wireless
Real-Time Video Transmission System[C]//

In this paper, a brand-new AI-based network 2018 International Conference on Sensor
topology optimization system is proposed. By Networks and Signal Processing (SNSP). IEEE
analyzing future network traffic development Computer Society, 2018.
trends via a traffic forecast model and refining

© International Telecommunication Union, 2021 87