A Constrained Dynamic Evolutionary Algorithm with Adaptive Penalty Coefficient

  • Authors:

  • Bo Xiao;Danpin Yu;Lei Zhang;Xin Tian;Song Gao;Sanyou Zeng

  • Affiliations:

  • Dept. of Computer Science Technology Research Center for Space Science& Technology The State Key Laboratory of Geological Processes Mineral Resources, China University of Geosciences, Wuhan, China ...;Dept. of Computer Science Technology Research Center for Space Science& Technology The State Key Laboratory of Geological Processes Mineral Resources, China University of Geosciences, Wuhan, China ...;Dept. of Computer Science Technology Research Center for Space Science& Technology The State Key Laboratory of Geological Processes Mineral Resources, China University of Geosciences, Wuhan, China ...;Dept. of Computer Science Technology Research Center for Space Science& Technology The State Key Laboratory of Geological Processes Mineral Resources, China University of Geosciences, Wuhan, China ...;Dept. of Computer Science Technology Research Center for Space Science& Technology The State Key Laboratory of Geological Processes Mineral Resources, China University of Geosciences, Wuhan, China ...;Dept. of Computer Science Technology Research Center for Space Science& Technology The State Key Laboratory of Geological Processes Mineral Resources, China University of Geosciences, Wuhan, China ...

  • Venue:
  • HAIS '08 Proceedings of the 3rd international workshop on Hybrid Artificial Intelligence Systems
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper proposes a new evolutionary algorithm with adaptive penalty coefficient. Firstly, the crossover operator of the new algorithm searches a lower-dimensional neighbor of the parent points, so that the algorithm converges fast, especially for high-dimensional problems. Secondly, the violation values of all constraint functions and the value of the objective function are normalized, and therefore, only one penalty coefficient is needed in the scheme. The penalty coefficient is selected adaptively. It is not too big, so as the algorithm can converge fast, and it is not too small so as the algorithm can avoid local optimal as much as possible. Thirdly, the standard deviation of violation values of the constraint functions is added to the violation item in the penalty function, and therefore, the individuals in the population can evenly approach the feasible region from the infeasible space. We have used the 24 constrained benchmark problems to test the new algorithm. The experimental results show it works better than or competitive to a known effective algorithm [7]