Page 111 - 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

Zhouwei Gang completed the optimization rules for But the "(ultra-low) *3" TOP (topology optimization
overload and low-load links, and Lin Xi realized the strategy) is still the topology optimization of the
aggregation of low-load links with the same head link granularity, can it be further refined to the node
and tail nodes. granularity?
Overload link optimization scheme： Lin Xi introduced a greedy algorithm to improve the
optimization strategy. The core strategy is to refer
Calculate the utilization rate of all links and the to the index score when optimizing overloaded
average value of the entire network based on the links and low-load links, and always move the nodes
predicted node traffic, search for all overloaded to get the highest index score. In this way, the node
links to form a set, and select a link, as shown in the is moved to the optimal link.
red link on the left. Along the link sequence, each
node is judged according to rules, whether it can The calculation of the index score is mainly related
establish a link with other link nodes, and the to the link optimization ratio, the average, minimum,
results are formed into a set. According to the load and maximum value of link bandwidth utilization,
of the original link and the load of the new link, find and the ratio of secondary links and downstream
a pair of nodes from the set, disconnect them, and nodes in the link. Using the optimization strategy of
connect them to other links to form a new link. the greedy algorithm, the nodes are not randomly
moved to other links, but selectively moved to the
The low-load link optimization rules are similar and links that make the index score higher, and the
will not be elaborated. optimization granularity is refined from the link
level to the node level, thereby improving the
overall optimization effect, with the link
optimization ratio increased by 0.4. Although the
calculation time doubled, it did not exceed 15
minutes, and the optimization evaluation score
increased by about 10%.
At the same time, Lin Xi has tested the genetic
algorithm. For the topic of network topology
optimization, the genetic algorithm involves more
debugging such as parameter coding, selection,
Fig. 13 – Link adjustment crossover, mutation, etc. For example, topology
L stands for link, J stands for J node. Through the "(ultra optimization is difficult to map and transform in the
low)*3" TOP (optimization strategy refers to the parameter coding part, and it involves rule
calculation of three operations repeated three times): constraints during crossover, considering the entire
1. Optimize overloaded links genetic algorithm for topology optimization is
2. Optimize low-load links actually very complicated, finally only the greedy
3. Adjust the threshold and recalculate the overload algorithm is used.
link and the low-load link
The algorithm can complete topology optimization in
55.5 minutes. Lin Guo used Qunee (Web graphics
component solution) to complete the topology map.
The left picture of the figure below is the original
topology of City C, and the right picture is the optimized
topology. Green nodes indicate adjusted nodes.

Fig. 15 – Greedy algorithm

Fig. 14 – Node load reduction

© International Telecommunication Union, 2021 95