Evolutionary many-objective algorithm with improved growing neural gas and angle-penalized distance for irregular problems

Qinghua Gu,Dejun Pang,Qian Wang

DOI: https://doi.org/10.1007/s10489-023-04526-3

IF: 5.3

2023-03-21

Applied Intelligence

Abstract:For solving complex many-objective optimization problems, the many-objective evolutionary algorithms may require new distance measures and more enormous selection pressures to guide solutions to the Pareto front. In this case, solving irregular Pareto front problems with growing neural gas networks as reference vectors is a great challenge. This paper proposes an algorithm with improved growing neural gas and angle-penalized distance to overcome the difficulty: First, a new parameter, the proportion of the solutions relatively close to the input data, is introduced to the angle-penalty distance, boosting the influence on the entire environment selection and enhancing the effect for Pareto front learning. Second, the edge-acceleration strategy enhances the ability of growing neural gas to match Pareto fronts and adapt to rapidly changing input. Finally, by adjusting the nodes deletion strategy in the search procedure, more promising solutions may be maintained to optimize the solutions’ distribution. In experiments tackling problems with irregular Pareto fronts, the proposed algorithm performs competitively in contrast to the other six excellent algorithms.

computer science, artificial intelligence