基于机器学习算法的区域建筑负荷预测建模研究  

Research on Modeling of Regional Cooling and Heating Load Forecast Based on Machine Learning Algorithm

作　　者：步婷 范蕊[1] 孙可欣 周亿冰 史洁[1] BU Ting;FAN Rui;SUN Kexin;ZHOU Yibing;SHI Jie(Tongji University,Shanghai 200092,China;Shanghai Research Institute of Building Science,Shanghai 201108,China)

机构地区：[1]同济大学,上海200092 [2]上海建筑科学研究院,上海201108

出　　处：《建筑科学》2022年第4期85-96,共12页Building Science

基　　金：国家自然科学基金项目“基于不确定性的集成低品位能源总线系统运行机理与多目标优化研究”(52078356)。

摘　　要：在全球能源转型和低碳可持续发展的趋势下,半集中式的区域供冷供热系统受到了广泛关注,而区域建筑冷热负荷预测是区域供冷供热系统优化控制的重要依据。本文对比了5种常见机器学习算法在不同输入组合下的商务区建筑负荷预测建模结果,提出了参数寻优和分段建模的优化方法,并引入区间和概率形式量化负荷预测的不确定性。研究结果表明,以温度和历史负荷为输入的随机森林预测模型准确度高稳定性好。文章的最后用实测数据对优化建模方法进行了验证性研究,证明了基于随机森林的负荷预测分段模型比多项式回归预测模型的准确度高实用性更强,并给出了实例模型预测的不确定性区间。Under the trend of global energy transition and low-carbon sustainable development, semi-centralized regional cooling and heating systems have received widespread attention. Regional load forecasting is an important basis for the optimal control of regional cooling and heating systems. By comparing the business district load forecast modeling results of five common machine learning algorithms under different input combinations, this paper proposes optimization methods for parameter optimization and segmented modeling, and introduces interval and probability forms to quantify the uncertainty of load forecasting. The research results show that the random forest load forecasting model with temperature and historical load as input has high accuracy and good stability. In the last part, a confirmatory study was conducted on the modeling method proposed in the article with measured data, which proved that the segmented load forecasting model based on the random forest is more accurate and more practical than the polynomial regression load forecasting model. And the uncertainty interval of the example model prediction is given.

关 键 词：负荷预测 机器学习 不确定性 优化建模 区域供冷供热 

分 类 号：TU831[建筑科学—供热、供燃气、通风及空调工程]