基于多代理辅助多目标进化优化的建筑节能智能设计方法  

Intelligent design of building energy conservation with multi-surrogate assisted MOEA

作　　者：张勇 梁晓珂 陈志鹏 巩敦卫 ZHANG Yong;LIANG Xiao-ke;CHEN Zhi-peng;GONG Dun-wei(School of Information and Control Engineering,China University of Mining and Technology,Xuzhou 221116,China)

机构地区：[1]中国矿业大学信息与控制工程学院,江苏徐州221116

出　　处：《控制与决策》2023年第11期3057-3065,共9页Control and Decision

基　　金：徐州市重点研发计划项目(KC20184);国家自然科学基金项目(62133015)。

摘　　要：进化优化具有优异的全局搜索能力,已成功应用于建筑节能设计问题.然而,由于需要借助代价高昂的建筑能耗软件不断评价个体,现有建筑节能设计进化算法普遍存在运行代价高的问题.鉴于此,提出一种面向建筑节能设计的多代理辅助多目标进化优化算法,简称MS-MOEA/D.首先,依据MOEA/D的目标分解特征同时构建多个基础代理模型;然后,针对每个待评估个体,自动选择合适的基础代理模型,并使用它们的集成结果预测该个体的目标值,达到提高其预测精度的目的.同时,在进化过程中自主确定基础代理模型的更新时机和规模,以降低代理模型的管理成本;最后,将所提出MS-MOEA/D与建筑能耗模拟软件EnergyPlus相融合,建立面向建筑节能设计的多目标进化优化仿真平台,并将该平台应用于中国北京地区常见居民和办公建筑节能设计实例中.通过与7种典型多目标进化算法进行对比,结果表明,MS-MOEA/D在显著降低计算代价的基础上能够得到高竞争力的Pareto最优解集.Evolutionary computation has been successfully applied to building energy conservation design problems because of excellent global search capabilities.However,since the need to continuously evaluate individuals by expensive building energy consumption software,existing evolutionary algorithms generally suffer from high operating cost.In view of this,this paper proposes a multi-surrogate assisted multi-objective evolutionary algorithm for building energy conservation design,called MS-MOEA/D.Firstly,multiple basic surrogate models are constructed simultaneously according to the objective decomposition characteristic of the multi-objective evolutionary algorithm MOEA/D.Then,for each individual that needs to be evaluated,the appropriate base surrogate model is automatically selected,and their integration results are used to predict the objective value of the individual,so as to improve its prediction accuracy.At the same time,the update timing and the scale of the basic surrogate model are determined autonomously in the evolution process,in order to reduce the management cost of the surrogate model.The MS-MOEA/D is integrated with the software EnergyPlus to establish a multi-objective evolutionary simulation platform for building energy conservation design,and the platform is used in the energy conservation design examples of atypical residential and office buildings in Beijing,China.Comparing with seven classical multi-objective evolutionary algorithms,experimental results show that the MS-MOEA/D can obtain a highly competitive Pareto optimal solution set while significantly reducing the computational cost.

关 键 词：进化优化 多目标 建筑节能 MS-MOEA/D 代理模型 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]