.png){kind=icon}
.png){kind=icon}
Return
A many-objective evolutionary algorithm with adaptive convergence calculation
.png){kind=icon}
PRE
AI
DOI：10.1007/s10489-022-04296-4.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Since different reference points are crucial for calculating convergence, we design a many-objective evolutionary algorithm with an adaptive convergence calculation method (ACC-MaOEA). This algorithm uses the adaptive convergence calculation method to estimate the shape of the Pareto front (PF) and adaptively determines a reference point to calculate convergence based on the shape. It estimates the PF shape by comparing the distances from the ideal and key points to two parallel planes. If the PF is concave, the ideal point is used as the reference point, and the distance from the solution to a plane through the ideal point is calculated to approximate convergence; if the PF is convex, the nadir point is used as the reference point, and the distance from the solution to a plane through the nadir point is calculated to approximate convergence. To avoid the overestimation of the nadir point, we first adopt a ratio-based infinite norm indicator to determine a potential region in which the optimal solution exists and then estimate the PF shape in this region and adaptively calculate convergence. Additionally, we use a determinantal point process to sample solutions with good convergence and diversity. We compare ACC-MaOEA with state-of-the-art algorithms on 21 test problems and up to 15 objectives. The experimental results show that ACC-MaOEA significantly outperforms its competitors, especially on regular PF problems.
Keywords:
Many-objective optimization
Evolutionary algorithm
Convergence
Pareto front estimate