Please wait a minute...
浙江大学学报(工学版)  2022, Vol. 56 Issue (5): 930-937    DOI: 10.3785/j.issn.1008-973X.2022.05.010
土木工程     
基于加权残差聚类的建筑负荷预测区间估计
章超波1,2,3(),刘永政4,李宏波1,2,*(),赵阳3,张丽珠3,王子豪3
1. 空调设备与系统节能国家重点实验室,广东 珠海 519000
2. 广东省制冷设备与节能技术重点实验室,广东 珠海 519000
3. 浙江大学 制冷与低温研究所,浙江 杭州 310027
4. 浙江大学 能源工程学院,浙江 杭州 310027
Weighted residual clustering-based building load prediction interval estimation
Chao-bo ZHANG1,2,3(),Yong-zheng LIU4,Hong-bo LI1,2,*(),Yang ZHAO3,Li-zhu ZHANG3,Zi-hao WANG3
1. State Key Laboratory of Air-Conditioning Equipment and System Energy Conservation, Zhuhai 519000, China
2. Guangdong Key Laboratory of Refrigeration Equipment and Energy Conservation Technology, Zhuhai 519000, China
3. Institute of Refrigeration and Cryogenics, Zhejiang University, Hangzhou 310027, China
4. College of Energy Engineering, Zhejiang University, Hangzhou 310027, China
 全文: PDF(1001 KB)   HTML
摘要:

提出基于加权残差聚类的建筑负荷预测区间估计方法,旨在对建筑负荷预测模型的不确定性进行定量评估. 使用Shapley additive explanations方法量化负荷预测模型的每个输入对输出的贡献程度. 基于得到的贡献程度对模型输入进行加权聚类,获得不同聚类簇中的模型历史残差分布. 根据不同聚类簇中的残差分布估计模型的预测区间. 在深圳某办公建筑1 a的冷负荷数据集上进行验证. 结果表明,与传统不对输入进行加权的方法相比,该方法可以显著提高预测区间的估计精度. 期望得到的预测区间与该方法得到的预测区间的平均覆盖误差为1.87%,而传统方法的平均覆盖误差为2.27%. 该方法可以用于估计任何数据驱动的建筑负荷预测模型的不确定性,从而为优化控制和故障诊断提供更可靠的负荷预测模型.

关键词: 建筑负荷预测区间估计数据驱动模型模型可解释性残差聚类    
Abstract:

A weighted residual clustering-based prediction interval estimation method was proposed for quantifying uncertainties in building energy load prediction. Firstly, the Shapley additive explanations approach was introduced to calculate a contribution level of each model input to model outputs for a specific load prediction model. Then, the contribution level was adopted for weighted clustering of model inputs to obtain the distribution of historical model residuals in different clusters. Finally, load prediction intervals of the model were estimated based on the distribution of residuals in different clusters. This method was validated on the one-year cooling load data set from a public building located in Shenzhen, Guangdong, China. Results showed that this method had higher accuracy of interval estimation than conventional methods whose inputs were not weighted in residual clustering. The average coverage error of prediction intervals was 1.87% using this method, while the average coverage error of prediction intervals was 2.27% using conventional methods. This method is applicable for any data-driven building energy load prediction models. It can be utilized to provide accurate and reliable building load prediction for optimal control and fault detection.

Key words: building energy load prediction    interval estimation    data-driven model    model interpretability    residual clustering
收稿日期: 2021-06-15 出版日期: 2022-05-31
CLC:  TU 831.2  
基金资助: 国家自然科学基金资助项目(51978601);空调设备与系统节能国家重点实验室资助项目(ACSKL2019KT07)
通讯作者: 李宏波     E-mail: chaoboo.zhang@zju.edu.cn;lihongbo@cn.gree.com
作者简介: 章超波(1994—),男,博士生,从事建筑大数据分析研究. oricid.org/0000-0001-6005-1051. E-mail: chaoboo.zhang@zju.edu.cn
服务  
把本文推荐给朋友
加入引用管理器
E-mail Alert
作者相关文章  
章超波
刘永政
李宏波
赵阳
张丽珠
王子豪

引用本文:

章超波,刘永政,李宏波,赵阳,张丽珠,王子豪. 基于加权残差聚类的建筑负荷预测区间估计[J]. 浙江大学学报(工学版), 2022, 56(5): 930-937.

Chao-bo ZHANG,Yong-zheng LIU,Hong-bo LI,Yang ZHAO,Li-zhu ZHANG,Zi-hao WANG. Weighted residual clustering-based building load prediction interval estimation. Journal of ZheJiang University (Engineering Science), 2022, 56(5): 930-937.

链接本文:

https://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2022.05.010        https://www.zjujournals.com/eng/CN/Y2022/V56/I5/930

图 1  预测区间估计方法流程图
超参数 寻优范围 最优值
dmax 2.0, 3.0, 4.0, 5.0 5.0
l 0.2, 0.4, 0.6, 0.8, 1.0 0.2
p1 0.2, 0.4, 0.6, 0.8, 1.0 0.8
p2 0.2, 0.4, 0.6, 0.8, 1.0 1.0
Ntree 50.0, 100.0, 150.0, 200.0 100.0
表 1  XGBoost超参数寻优结果
数据集 MAE/kW RMSE/kW R2 CV-RMSE/%
训练集 321.44 531.52 0.94 15.93
测试集 366.15 616.83 0.92 15.74
表 2  XGBoost模型在训练集和测试集上的精度
图 2  模型输入的局部SHAP值小提琴图
变量 $\bar{\phi }$ I
M 0.04 0.02
H 0.13 0.04
W 0.04 0.01
Tout 0.23 0.08
RHout 0.01 0.01
CL1 0.39 0.54
CL2 0.12 0.30
表 3  模型输入的归一化全局SHAP值和XGBoost特征重要性
k $\left| { \overline{ {\rm{ACE} } } } \right|$/%
本研究方法 传统方法
2 2.37 2.77
3 1.87 2.63
4 2.08 2.93
5 2.10 2.71
6 2.23 2.58
7 2.50 2.64
8 2.34 2.40
9 2.48 2.41
10 2.82 2.27
表 4  不同k值下的ACE绝对值的平均值
图 3  k = 3时残差簇的残差分布图
PINC/% PICP/% ACE/%
10 9.75 ?0.25
20 19.55 ?0.45
30 29.33 ?0.67
40 38.36 ?1.64
50 47.78 ?2.22
60 57.11 ?2.89
70 67.17 ?2.83
80 76.99 ?3.01
90 87.11 ?2.89
表 5  不同名义置信水平下的预测区间估计性能(k = 3)
图 4  某典型日的实际负荷曲线、预测负荷曲线及估计得到的预测区间(PINC=80%、90%)
1 龙惟定, 梁浩 我国城市建筑碳达峰与碳中和路径探讨[J]. 暖通空调, 2021, 51 (4): 1- 17
LONG Wei-ding, LIANG Hao Discussion on paths of carbon peak and carbon neutrality of urban buildings in China[J]. Journal of HV and AC, 2021, 51 (4): 1- 17
2 清华大学建筑节能研究中心. 中国建筑节能年度发展研究报告2020[M]. 北京: 中国建筑工业出版社, 2020.
3 刘奇特. 基于群智能的中央空调冷站节能优化方法研究[D]. 西安: 西安建筑科技大学, 2020.
LIU Qi-te. Research on the energy-saving optimization methods of central air-conditioning chilled station based on the insect intelligence [D]. Xi’an: Xi’an University of Architecture and Technology, 2020.
4 ZHANG C, LI J, ZHAO Y, et al A hybrid deep learning-based method for short-term building energy load prediction combined with an interpretation process[J]. Energy and Buildings, 2020, 225: 110301
doi: 10.1016/j.enbuild.2020.110301
5 GASSAR A A A, CHA S H Energy prediction techniques for large-scale buildings towards a sustainable built environment: a review[J]. Energy and Buildings, 2020, 224: 110238
doi: 10.1016/j.enbuild.2020.110238
6 孙超, 孟祥萍, 王晖, 等 改进BP神经网络的楼宇负荷预测[J]. 自动化应用, 2020, (10): 61- 64
SUN Chao, MENG Xiang-ping, WANG Hui, et al Research on building load forecasting based on improved BP neural network[J]. Automation Application, 2020, (10): 61- 64
7 FAN C, XIAO F, ZHAO Y A short-term building cooling load prediction method using deep learning algorithms[J]. Applied Energy, 2017, 195: 222- 233
doi: 10.1016/j.apenergy.2017.03.064
8 SAEEDI M, MORADI M, HOSSEINI M, et al Robust optimization based optimal chiller loading under cooling demand uncertainty[J]. Applied Thermal Engineering, 2019, 148: 1081- 1091
doi: 10.1016/j.applthermaleng.2018.11.122
9 ZHANG C, ZHAO Y, FAN C, et al A generic prediction interval estimation method for quantifying the uncertainties in ultra-short-term building cooling load prediction[J]. Applied Thermal Engineering, 2020, 173: 115261
doi: 10.1016/j.applthermaleng.2020.115261
10 DAHL M, BRUN A, ANDRESEN G B Using ensemble weather predictions in district heating operation and load forecasting[J]. Applied Energy, 2017, 193: 455- 465
doi: 10.1016/j.apenergy.2017.02.066
11 RODRÍGUEZ F, BAZMOHAMMADI N, GUERRERO J M, et al A very short-term probabilistic prediction interval forecaster for reducing load uncertainty level in smart grids[J]. Applied Sciences, 2021, 11 (6): 2538
doi: 10.3390/app11062538
12 GUAN C, LUH P B, MICHEL L D, et al Hybrid Kalman filters for very short-term load forecasting and prediction interval estimation[J]. IEEE Transactions on Power Systems, 2013, 28 (4): 3806- 3817
doi: 10.1109/TPWRS.2013.2264488
13 ZHANG C, ZHAO Y, ZHANG X, et al An improved cooling load prediction method for buildings with the estimation of prediction intervals[J]. Procedia Engineering, 2017, 205: 2422- 2428
doi: 10.1016/j.proeng.2017.09.967
14 QUAN H, SRINIVASAN D, KHOSRAVI A Uncertainty handling using neural network-based prediction intervals for electrical load forecasting[J]. Energy, 2014, 73: 916- 925
doi: 10.1016/j.energy.2014.06.104
15 WANG J, GAO Y, CHEN X A novel hybrid interval prediction approach based on modified lower upper bound estimation in combination with multi-objective salp swarm algorithm for short-term load forecasting[J]. Energies, 2018, 11 (6): 1561
doi: 10.3390/en11061561
16 XU C, CHEN H A hybrid data mining approach for anomaly detection and evaluation in residential buildings energy data[J]. Energy and Buildings, 2020, 215: 109864
doi: 10.1016/j.enbuild.2020.109864
17 WANG L, KUBICHEK R, ZHOU X Adaptive learning based data-driven models for predicting hourly building energy use[J]. Energy and Buildings, 2018, 159: 454- 461
doi: 10.1016/j.enbuild.2017.10.054
18 WANG Z, HONG T, PIETTE M A Building thermal load prediction through shallow machine learning and deep learning[J]. Applied Energy, 2020, 263: 114683
doi: 10.1016/j.apenergy.2020.114683
19 DING Y, ZHANG Q, YUAN T, et al Effect of input variables on cooling load prediction accuracy of an office building[J]. Applied Thermal Engineering, 2018, 128: 225- 234
doi: 10.1016/j.applthermaleng.2017.09.007
20 ARJUNAN P, POOLLA K, MILLER C EnergyStar++: towards more accurate and explanatory building energy benchmarking[J]. Applied Energy, 2020, 276: 115413
doi: 10.1016/j.apenergy.2020.115413
21 FAN C, XIAO F, YAN C, et al A novel methodology to explain and evaluate data-driven building energy performance models based on interpretable machine learning[J]. Applied Energy, 2019, 235: 1551- 1560
doi: 10.1016/j.apenergy.2018.11.081
22 FRÄMLING K, WESTBERG M, JULLUM M, et al. Comparison of contextual importance and utility with LIME and Shapley values [C]// Proceedings of the 3rd International Workshop on Explainable, Transparent Autonomous Agents and Multi-Agent Systems. Cham: Springer, 2021: 39–54.
23 LUNDBERG S M, LEE S I. A unified approach to interpreting model predictions [C]// Proceedings of the 31st International Conference on Neural Information Processing Systems. New York: Association for Computing Machinery, 2017: 4768–4777.
24 ZHANG C, XUE X, ZHAO Y, et al An improved association rule mining-based method for revealing operational problems of building heating, ventilation and air conditioning (HVAC) systems[J]. Applied Energy, 2019, 253: 113492
doi: 10.1016/j.apenergy.2019.113492
25 ZHANG L, WEN J A systematic feature selection procedure for short-term data-driven building energy forecasting model development[J]. Energy and Buildings, 2019, 183: 428- 442
doi: 10.1016/j.enbuild.2018.11.010
No related articles found!