Study Of Multi-objective Evolutionary Algorithms Based On Preference Information And Related Indicator

Most real-life problems usually involve multiple conflicting objectives and that needs to be optimized simultaneously under certain constraints.Such problems are known as multi-objective optimization problems(MOPs),which hard to solved by traditional optimization methods.Multi-objective evolutionary algorithms(MOEAs)have been widely adopted to deal with such problems since it can obtain a set of Pareto-optimal solutions that well approximate the entire Pareto-optimal front(POF)at the same time.In most case,what the DM interested in is solutions set that obtained in a partial region,which named the region of interest(ROI).Thus,preference-based MOEAs have been developed to facilitate the DM by providing solutions in the regions of interest(ROIs)pursuant to DMs preference information.By incorporating the DMs preferences or expert knowledge,the preference-based MOEAs can reduce the search space and reduce computation in the search process.Guided by the preference information,it largely reduces the time of solving process and ultimately provides the solutions that decision maker expect.The research work and innovation are as follows,aiming at the problems in the field of preference-based multi-objective evolutionary algorithms,this paper proposes a solution which converting user-preference into constraints in multi-objective optimization.The basic idea of the proposed method is to convert user preference information into constraints which is called as preference constraints.Then the excellent MOEAs combined constraint handling techniques can solve the MOPs with preference constraints.The proposed approach has the advantage of being very easy to couple into any MOP,This kind of convert is independent of the MOEA used and can be easily coupled to any of them without any modification to the main structure of the method chosen.The process of our proposed method is simple in principle and can meet the decision maker's requirements.In the other hand,aiming at the difficulty in evaluating preference-based evolutionary multi-objective optimization,this paper proposes a new performance indicator.The main advantage of indicator is not need to obtain true POF of multi-objective problem The main idea is to project the preferred solutions onto a constructed hyper plane which is perpendicular to the vector from the reference(aspiration)point to the origin.And then the distance from preferred solutions to the origin and the standard deviation of distance from each mapping point to the nearest point will be calculated.The former is used to measure the convergence of the obtained solutions.The latter is utilized to assess the diversity of preferred solution the region of interest.The indicator is conducted to assess different algorithms on a series of benchmark problems with various features.The results show that the proposed indicator is able to properly evaluate the performance of preference-based multi-objective evolutionary algorithms.