TY - CHAP
T1 - Reinforcement learning based NSGA-II for energy-delay trade-off in IAB mmWave Het-Nets
AU - Shang, Wen
AU - Friderikos, Vasilis
N1 - Publisher Copyright:
© 2023 IEEE.
PY - 2023
Y1 - 2023
N2 - This paper proposes efficient ways for constructing the energy-delay Pareto front in cache-enabled integrated access and backhauling (IAB) heterogeneous wireless network. More specifically, an improved non-dominated sorting genetic algorithm II (NSGA-II) is proposed which is coupled with the operator parameter control ability based on reinforcement learning, to solve the energy-delay trade-off in the multi-objective optimization problem. To estimate the effectiveness of the proposed scheme, key performance indicators that cover the convergence and distribution of the Pareto front solution set are conducted and analyzed. A wide set of numerical investigations show that the proposed algorithm can provide a more evenly distributed result than the state of the art techniques with a 15% gain compared to the nominal case which is the weighted-sum method. Furthermore, and maybe more importantly, the undesirable large gaps between solutions in the Pareto front which are caused by the weighting coefficient choices are avoided. By enabling the operator parameter control ability, the exploration and exploitation process of the proposed algorithm can be balanced, which prevents the frequently faced problems of early convergence and being trapped at a local optimum in the genetic algorithm. The proposed technique can have significant implications in improving the avoidable choices regarding the network operation, and compared with the traditional NSGA-II, the proposed algorithm can provide a near-optimal solution set with 20% more diversity.
AB - This paper proposes efficient ways for constructing the energy-delay Pareto front in cache-enabled integrated access and backhauling (IAB) heterogeneous wireless network. More specifically, an improved non-dominated sorting genetic algorithm II (NSGA-II) is proposed which is coupled with the operator parameter control ability based on reinforcement learning, to solve the energy-delay trade-off in the multi-objective optimization problem. To estimate the effectiveness of the proposed scheme, key performance indicators that cover the convergence and distribution of the Pareto front solution set are conducted and analyzed. A wide set of numerical investigations show that the proposed algorithm can provide a more evenly distributed result than the state of the art techniques with a 15% gain compared to the nominal case which is the weighted-sum method. Furthermore, and maybe more importantly, the undesirable large gaps between solutions in the Pareto front which are caused by the weighting coefficient choices are avoided. By enabling the operator parameter control ability, the exploration and exploitation process of the proposed algorithm can be balanced, which prevents the frequently faced problems of early convergence and being trapped at a local optimum in the genetic algorithm. The proposed technique can have significant implications in improving the avoidable choices regarding the network operation, and compared with the traditional NSGA-II, the proposed algorithm can provide a near-optimal solution set with 20% more diversity.
UR - http://www.scopus.com/inward/record.url?scp=85178280411&partnerID=8YFLogxK
U2 - 10.1109/ICC45041.2023.10278969
DO - 10.1109/ICC45041.2023.10278969
M3 - Conference paper
T3 - IEEE International Conference on Communications
SP - 4206
EP - 4211
BT - ICC 2023 - IEEE International Conference on Communications
A2 - Zorzi, Michele
A2 - Tao, Meixia
A2 - Saad, Walid
ER -