基于搜索过程知识表示与推理的进化多目标优化算法研究

项目名称： 基于搜索过程知识表示与推理的进化多目标优化算法研究

项目编号： No.61303119

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 戚玉涛

作者单位： 西安电子科技大学

项目金额： 25万元

中文摘要： 进化多目标优化（EMO）算法是目前求解多目标优化问题（MOPs）的主流方法之一。然而，EMO算法搜索过程的盲目性导致这类算法需花费大量函数评价代价才能收敛。针对该瓶颈问题，本课题拟研究EMO算法中搜索过程知识表示和搜索方向推理方法，用于提高算法的搜索效率。首先，建立求解MOPs的多智能体网络模型，给出基于此模型的EMO算法框架。模型将MOPs转化为一组单目标优化子问题分配给各个智能体求解，算法采用集成学习方法将多个智能体训练的弱Surrogate模型合成泛化能力强的集成模型逼近MOPs目标函数，并用于估计新解的质量，以降低算法函数评价代价。然后，设计EMO算法在决策空间和目标空间上搜索过程知识的表示和学习方法，给出两类知识的相互转化机制，提出基于两类知识的搜索方向推理方法，以降低搜索过程的盲目性。最后，在理论研究基础上，开展对水库防洪调度问题的应用研究，进一步推动EMO算法在工程中的应用。

中文关键词： 进化计算；多目标优化；机器学习；搜索过程知识；水库防洪调度

英文摘要： Evolutionary multi-objective optimization (EMO) algorithm is one of the mainstream approaches to multi-objective optimization problems (MOPs). However, EMO algorithm spends a large amount of function evaluation effort before convergence, due to the blindness of its searching process. To remedy this, this proposal focuses on the key techniques of the knowledge representation and the searching direction reasoning in EMO algorithm, with the help of which the efficiency of the searching process can be improved. First, a multi-agent network model will be build for solving MOPs. Based on the model, a new framework of EMO algorithm will be developed. The model transforms MOPs into a set of scalar optimization subproblems and assigns these optimization subtasks to the agents. Using the ensemble learning method, the new EMO algorithm combines several weak surrogate models trained by the agents into an ensemble surrogate model. The ensemble surrogate model is then used to estimate the quality of new solutions. It helps to reduce the function evaluation costs of the EMO algorithm. Second, the representation and learning method of the searching process knowledge in both the decision space and the objective space will be designed. A transformation method will also be developed to convert these two types of knowledge into eac

英文关键词： Evolutionary computation；Multi-objective optimization；Machine learning；Knowledge in search procedure；Reservoir flood control operation