Improved Interval Multi-objective
Evolutionary Optimization Algorithm
Based on Directed Graph
Xiaoyan Sun, Pengfei Zhang
(&)
, Yang Chen, and Yong Zhang
School of Information and Control Engineering,
China University of Mining and Technology, Xuzhou 221116, China
zhangpongfy@sina.com
Abstract. Multi-objective evolutionary algorithm for optimizing objectives
with interval parameters is becoming more and more important in practice. The
efficient comparison metrics on interval values and the associated offspring
generations are critical. We first present a neighboring dominance metric for
interval numbers comparisons. Then, the potential dominant solutions are pre-
dicted by constructing a directed graph with the neighboring dominance. We
design a directed graph using those competitive solutions sorted with NSGA-II,
and predict the possible evolutionary paths of next generation. A PSO mechanic
is applied to generate the potential outstanding solutions in the paths, and these
solutions are further used to improve the crossover efficiency. The experimental
results demonstrate the performance of the proposed algorithm in improving the
convergence of interval multi-objective evolutionary optimization.
Keywords: Interval multi-objective
Evolutionary optimization Directed
graph
1 Introduction
Multi-objective evolutionary algorithms (MOEAs) based on the concept of Pareto
dominance such as vector evaluation genetic algorithm [1], non-dominated sort genetic
algorithm (NSGA) [2], improved NSGA-II [3] and multi-objective evolutionary opti-
mization algorithm based on decomposition [4] are powerful for solving
multi-objective problems [5]. In practice, numerous multi-objective optimization
problems encounter uncertain parameters.
When the parameters are intervals, the values of the objectives are intervals too,
which are called interval multi-objective optimization problems. Aiming at solving the
multi-objective optimization problems with the objective functions having noise,
Limbourgetc. used the interval to represent the uncertain objectives, defined the interval
vector partial order relationship and the hyper-volume to measure the distribution and
approximation of the optimized solutions, and then proposed the evolutionary opti-
mization method, i.e., Imprecision Propagating Multi-Objective Evolutionary Algo-
rithm(IP-MOEA) [6]. Eskandarietc. proposed a stochastic Pareto genetic algorithm
(SPGA) to solve the above similar problems [7]. In addition, Gong et al. proposed
© Springer International Publishing AG 2017
Y. Tan et al. (Eds.): ICSI 2017, Part II, LNCS 10386, pp. 40–48, 2017.
DOI: 10.1007/978-3-319-61833-3_5