分类
多目标优化
数学优化
计算机科学
超参数
遗传算法
灵敏度(控制系统)
集合(抽象数据类型)
采样(信号处理)
过程(计算)
最优化问题
方案(数学)
人工神经网络
帕累托原理
全局优化
算法
工程类
数学
机器学习
程序设计语言
数学分析
操作系统
滤波器(信号处理)
计算机视觉
电子工程
作者
Ruijun Chen,Yaw-Shyan Tsay,Shiwen Ni
标识
DOI:10.1016/j.jclepro.2022.131978
摘要
In this paper, we proposed an integrated optimization framework to explore minimum building carbon emissions (BCE), indoor discomfort hours (IDH), and global cost (GC) of building, as well as a new formula for selecting the best scheme in the Pareto front set. Such framework improves the efficiency of the optimization process, the accuracy of the results, and the rationality of the best scheme. The entire optimization process can be divided into four steps. First, the input parameters were randomly generated using three sampling methods and then simulated to build the database. Then, the contribution rates of the selected parameters to the outputs were comprehensively evaluated and combined with multi-sensitivity analysis methods to screen important parameters. Next, we trained and validated the Back Propagation Neural Network (BPNN) model, in which different methods were used for hyperparametric optimization. Third, based on the comparison of various optimization methods, the Non-dominated Sorting Genetic Algorithm-III (NSGA-III) was selected and combined with BPNN to solve the proposed multi-objective optimization problem. Then, finally, we applied the proposed optimal balance formula to select the scheme that considered all aspects of objectives in the Pareto front set. The results demonstrated that the best sampling method and hyperparameter combination can result in an R2 of BPNN that reaches 0.992. The simulation results are in good agreement with the optimization results. Compared with the case building, the optimal balance schemes of BCE, IDH, and GC were reduced by 53.25%, 42.95%, and 22.33%, respectively. Therefore, we demonstrated that this method is feasible and effective for improving building design in more practical and complex situations and can be widely popularized in the building performance optimization field.
科研通智能强力驱动
Strongly Powered by AbleSci AI