摘要
提升建筑负荷预测模型的准确性和泛化能力,对于优化暖通空调系统运行策略、降低建筑能耗具有重要的指导作用。精准预测建筑热负荷已成为建筑节能研究中的核心问题之一。本文提出了一种基于遗传算法(GA)与堆叠(Stacking)算法的集成算法的建筑热负荷预测方法,根据现场调研数据和模拟仿真结果建立数据集,利用最大相关最小冗余(MRMR)与主成分分析(PCA)算法确定输入参数并验证其有效性。该数据处理流程为后续建模与分析提供了可靠的数据支持。在此基础上,分析了12种经典单一模型的负荷预测精度和拟合度,提出了一种基于GA-Stacking集成算法的建筑热负荷预测方法。其中,GA算法用于优化模型选择和参数调整,Stacking算法则通过结合多个基模型的输出提升预测性能。利用该集成算法对天津市某图书馆的采暖季热负荷进行了预测,并与实测数据进行了对比。结果表明:相较于12种单一算法,GA-Stacking集成算法的建筑热负荷预测结果的平均绝对误差(MAE)值降低了52.52 kW·h,均方误差(MSE)值下降了21 807.19(kW·h)^(2),均方根误差变异系数(CV)值降低了6.98%,决定系数(R^(2))提高了4.2%;相较于Stacking算法,MAE值降低了13.11 kW·h,MSE值减少了3 525.62(kW·h)^(2),CV值降低了1.57%,R^(2)提升了0.68%。经过模型筛选后的集成算法在预测精度、泛化能力和稳定性方面均表现更优。
Improving the accuracy and generalization ability of building load forecasting models plays an im⁃portant guiding role in optimizing the operation strategy of heating ventilating and air conditioning(HVAC)systems and reducing building energy consumption.Precise thermal load prediction has become a critical as⁃pect of building energy efficiency research.A thermal load prediction method based on an ensemble algorithm that integrates genetic algorithm(GA)and Stacking was proposed in this paper.A dataset was constructed us⁃ing field survey data and simulation results,and input parameters were selected and validated using the maxi⁃mum relevance minimum redundancy(MRMR)and principal component analysis(PCA).This data process⁃ing framework provides reliable support for subsequent modeling and analysis.Based on this framework,the prediction accuracy and fitting performance of 12 classical single models were evaluated,and a GA⁃Stacking ensemble algorithm for building thermal load prediction was proposed.The GA algorithm was employed for model selection and parameter optimization,while the Stacking algorithm enhanced prediction performance by integrating outputs from multiple base models.The proposed ensemble algorithm was applied to predict the heating season thermal load of a library in Tianjin,and the results were compared with measured data.The re⁃sults show that compared with the 12 single models,the GA⁃Stacking model reduces the mean absolute error(MAE)by 52.52 kW·h,decreases the mean squared error(MSE)by 21807.19(kW·h)^(2),lowers the coeffi⁃cient of variation of the root mean squared error(CV)by 6.98%,and improves the coefficient of determination(R^(2))by 4.2%.Relative to the Stacking model,the GA⁃Stacking model decreases MAE by 13.11 kW·h,re⁃duces MSE by 3525.62(kW·h)^(2),lowers CV by 1.57%,and increases R^(2)by 0.68%.The ensemble algorithm selected through model screening outperforms others in prediction accuracy,generalization,and stability.
作者
马洪亭
马小宇
杨开元
杨利利
王科荀
MA Hongting;MA Xiaoyu;YANG Kaiyuan;YANG Lili;WANG Kexun(School of Environmental Science and Engineering,Tianjin University,Tianjin 300354,China;Xinzheng Urban Development Investment Co.,Ltd.,Zhengzhou 451100,China;Zhonghe Gangda Construction Consulting Co.,Ltd.,Zhengzhou 451162,China)
出处
《东南大学学报(自然科学版)》
北大核心
2025年第6期1670-1682,共13页
Journal of Southeast University:Natural Science Edition
基金
国家自然科学基金青年基金资助项目(52208120).
关键词
遗传算法
堆叠算法
集成算法
负荷预测
泛化能力
genetic algorithm
stacking algorithm
integrated algorithm
load forecasting
generalization ability
