Integrated Optimization on Energy Saving and Quality of Service of Urban Rail Transit System
Table 6
Comparative analysis of several major multi-objective optimization algorithms.
Algorithm
Advantage
Insufficient
NSGA
The number of optimization objectives is unrestricted, the distribution of noninferior optimal solutions is uniform, and multiple equivalent solutions are allowed to exist.
The computational efficiency is low, and the shared parameters should be determined in advance.
PAES
Using “1 + 1” strategy and local search evolutionary strategy, making its solution time lower than other algorithms.
Easy to lose the horizontal or vertical Pareto front solution.
SPEA
Using an external population to realize elite retention strategy.
Using clustering to delete individuals from the external population, which may lose the noninferior solution in the external population.
MOEA/D
The convergence rate is faster, and the computational complexity is lower. Because the weight vectors guiding evolution are uniformly distributed, the solutions obtained by MOEA/D are uniformly distributed.
When dealing with multi-objective optimization problems with high dimensions, its distribution cannot be guaranteed, and the effect is poor.
NSGA-II
Using the crowded-comparison operator and elite strategy to expand the sampling space, which allows the parents and their offspring to participate in the competition to produce the next generation of the population and generate better offspring.
When dealing with multi-objective optimization problems, congestion distance is not applicable in high-dimensional space, and computational complexity is high.
