Journal
APPLIED INTELLIGENCE
Volume 52, Issue 11, Pages 12888-12923Publisher
SPRINGER
DOI: 10.1007/s10489-021-03003-z
Keywords
Differential evolution (DE); Niching; Multimodal optimization problem (MOP); Population topology; Self-adaptive
Categories
Funding
- fund of Research on Intelligent Ship Testing and Verification [[2018]473]
- Natural Science Foundation of China [51709027]
Ask authors/readers for more resources
In this paper, a Self-adaptive Niching Differential Evolution (SaNDE) algorithm is proposed to solve multi-modal optimization problems using a ring topology and local memory. The algorithm demonstrates competitiveness and superiority through innovative strategies such as self-adaptive control parameters and an adaptive re-start mechanism using opposition-based learning.
To solve multi-modal optimization problems, the niching technique is widely used because it could find and preserve multiple stable sub-populations. However, the performances of most existing evolutionary algorithms with niching techniques heavily depend on niching parameters, such as niche radius, sub-population size and crowding factor. To our best knowledge, a self-adaptive differential evolution (DE) variant without niching parameters using ring topology has not been developed. In this paper, we proposed a Self-adaptive Niching Differential Evolution (SaNDE) algorithm. The ring topology plays a crucial role in slowing the information flow, resulting in scattered niches with restricted and overlapped communications. We introduced local memory (personal best) into the DE algorithm to present a new mutation operator current-to-pnbest when a ring population topology was used. Moreover, the two control parameters in DE were self-adapted by using a simple but effective strategy that is based on successful parametric values in history. To improve the capability of jumping out of local optima, an adaptive re-start mechanism by using opposition-based learning was proposed to address the issue of stagnation. The performances of the proposed method were investigated through standard benchmark functions and the problem of optimizing parameters for a feedforward neural network. Comparisons with other state-of-the-art multi-modal optimization algorithms demonstrated the competitiveness of the proposed methodology.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available