Adaptive Multi-Objective Optimization Based on Feedback Design  

作　　者：窦立谦 宗群 吉月辉 曾凡琳 

机构地区：[1]School of Electrical Engineering and Automation,Tianjin University

出　　处：《Transactions of Tianjin University》2010年第5期359-365,共7页天津大学学报（英文版）

基　　金：Supported by National Natural Science Foundation of China (No.60874073);Tianjin Science and Technology Keystone Project (No.08ZCKFJC27900);Natural Science Foundation of Tianjin(No.08JCYBJC11900)

摘　　要：The problem of adaptive multi-objective optimization(AMOO) has received extensive attention due to its practical significance.An important issue in optimizing a multi-objective system is adjusting the weighting coefficients of multiple objectives so as to keep track of various conditions.In this paper,a feedback structure for AMOO is designed.Moreover,the reinforcement learning combined with hidden biasing information is applied to online tuning weighting coefficients of objective functions.Finally,the prop...The problem of adaptive multi-objective optimization(AMOO) has received extensive attention due to its practical significance.An important issue in optimizing a multi-objective system is adjusting the weighting coefficients of multiple objectives so as to keep track of various conditions.In this paper,a feedback structure for AMOO is designed.Moreover,the reinforcement learning combined with hidden biasing information is applied to online tuning weighting coefficients of objective functions.Finally,the proposed approach is applied to the optimization design problem of an elevator group control system.Simulation results show that AMOO has the best average performance at up-peak traffic profile,and its average waiting time reaches 22 s.AMOO is suitable for various traffic patterns,and it is also superior to the majority of algorithms at down-peak traffic profile.

关 键 词：multi-objective optimization adaptive optimization reinforcement learning elevator group system 

分 类 号：O221.6[理学—运筹学与控制论]