Knowledge-inducing MOEA/D for interval multi-objective optimization problems
Yinan Guo, Jianwei Cheng, Zhen Yang, Wang Chun · 2016
In practical engineering optimization problems, the parameters of the objective functions may be interval. Existing interval multi-objective optimization methods mostly adopted interval dominant relationships and corresponding crowded operators to find the better solutions. They were complex and time-consuming. So a novel knowledge-inducing interval MOEA/D is put forward. It decomposes interval multi-objective optimization problem to many interval single-objective optimization sub-problems in terms of the weight so as to realize the parallel exploration and avoid the complex dominant comparison. Especially, the improved Chebyshev aggregate function is defined to calculate the distance between the midpoint of the interval objective values and the reference point. Moreover, the reference point depends on the upper limit of the interval objective values. In order to keep the diversity of population and avoid the premature convergence, three kinds of knowledge including situational knowledge, neighborhood knowledge and association knowledge are defined and the differential evolution guided by above knowledge are illustrated in detail. The simulation results for five benchmark functions indicate that the proposed algorithm keeps the diversity of population better and obtains the Pareto-optimal solutions more close to the true Pareto front by comparing with other interval multi-objective optimization algorithms.