Privacy-Preserving Federated Learning for VRF System Energy Use Prediction with SMOGN Augmentation And GA-Based Aggregation

Zhipeng Guo; Bihai Wang; Yu Jiang; Leqi Zhu; Yixing Chen; Guanjing Lin

doi:10.59717/ipj.energy-use.2026.100060

Energy Use > 2026 Vol. 2 > No. 2 > 100060

Next Article Previous Article

ARTICLE Open Access

Cite PDF

Privacy-Preserving Federated Learning for VRF System Energy Use Prediction with SMOGN Augmentation And GA-Based Aggregation

1.
Department of mechanical and automation engineering, Chinese University of Hong Kong, Hong Kong SAR, China
2.
Institute of Future Human Habitats, Tsinghua Shenzhen International Graduate School, Tsinghua University, Shenzhen 518055, China
3.
Midea Building Technologies, Midea Group, Foshan 528311, China
4.
College of Civil Engineering, Hunan University, Changsha 410082, China
5.
Key Laboratory of Building Safety and Energy Efficiency of Ministry of Education, Hunan University, Changsha 410082, China
6.
These authors contributed equally

Show all affliationsShow less

Fund Project: This work was supported by scientific research start-up funds grant QD2024005C from Tsinghua Shenzhen International Graduate School, Tsinghua University. The data used in this study were provided by Midea Group, with permission granted for their use in this research.

More Information

Corresponding author: linguanjing@sz.tsinghua.edu.cn
Received Date: June 14, 2026

Accepted Date: June 23, 2026

Published Online: June 25, 2026

Energy Use 2(2), https://doi.org/10.59717/ipj.energy-use.2026.100060 | Cite this article

GRAPHICAL ABSTRACT

GRAPHICAL ABSTRACT

DownLoad: Full size image

PUBLIC SUMMARY

PUBLIC SUMMARY
1. Precise VRF energy prediction guides equipment matching and operation; traditional centralized forecasting risks user data privacy, while federated learning trains global models with local raw data retained.
  
  An improved federated learning framework combines client-side SMOGN data augmentation and GA weighted aggregation to solve VRF data imbalance and poor model fusion defects.
  
  Dataset of 518 Chinese VRF systems (3232 monthly records) is expanded to 4500 samples via SMOGN; three ML models are tested with building and equipment characteristic inputs.
  
  The proposed privacy-preserving FL framework reaches R=0.8390, approaching centralized benchmark R=0.8913, verifying its accuracy and scalability for VRF energy forecasting.

ABSTRACT

ABSTRACT

Accurate prediction of energy use for Variable Refrigerant Flow (VRF) systems is critical for optimizing equipment selection and efficient operation. Traditional prediction methods typically depend on centralized data collection, raising privacy concerns. Federated Learning (FL) provides a privacy-preserving alternative by enabling multiple clients to collaboratively train a global model while keeping raw data localized. However, practical FL applications for VRF systems face challenges of data imbalance and suboptimal model aggregation. To address these issues, this paper proposes an enhanced FL framework integrating Synthetic Minority Oversampling Technique for Regression with Gaussian Noise (SMOGN) and Genetic Algorithm (GA) to improve prediction accuracy while preserving privacy. SMOGN is employed at client level to augment local datasets and mitigate imbalance. A GA-based performance-weighted ensemble aggregation strategy is introduced to optimize model fusion in FL, replacing conventional uniform averaging and enabling better utilization of heterogeneous local models. The framework was evaluated using data from 518 VRF systems in China (3,232 monthly records, expanded to 4,500 samples via SMOGN). Three machine learning models, i.e. Artificial Neural Networks, Light Gradient Boosting Machine, and eXtreme Gradient Boosting, were implemented under centralized and federated settings, using five building features and nine equipment features as inputs to predict monthly energy consumption. Results show that the proposed FL framework achieves an R² value of 0.8390, close to centralized model performance (R² = 0.8913), while preserving data privacy. These findings demonstrate the effectiveness of the proposed approach as a scalable and privacy-preserving solution for energy consumption prediction in VRF systems.

FullText(HTML)

REFERENCES

[1]	Lin G., Casillas A., Sheng M., et al. (2023). Performance Evaluation of an Occupancy-Based HVAC Control System. Energies 16:7088. DOI:10.3390/en16207088 View in Article CrossRef Google Scholar
[2]	United Nations Environment Programme. (2024). 2023 Global Status Report for Buildings and Construction: Beyond Foundations – Mainstreaming Sustainable Solutions to Cut Emissions from the Buildings Sector. UNEP. DOI:10.59117/20.500.11822/45095. View in Article Google Scholar
[3]	National Development and Reform Commission, Ministry of Housing and Urban-Rural Development. (2022). 14th Five-Year Plan for Building Energy Efficiency and Green Building Development. China Architecture & Building Press. View in Article Google Scholar
[4]	Shin M., Kim S., Kim Y., et al. (2024). Development of an HVAC system control method using deep reinforcement learning. Build. Environ. 248:111069. DOI:10.1016/j.buildenv.2024.111069 View in Article CrossRef Google Scholar
[5]	Hagström F., Garg V., Oliveira F., et al. (2025). Employing federated learning for training autonomous HVAC systems. Energy Build. 340:115761. DOI:10.1016/j.enbuild.2025.115761 View in Article CrossRef Google Scholar
[6]	Enteria N., Sawachi T., Saito K., et al. (2023). Variable refrigerant flow systems: advances and applications. Springer Nature Singapore. DOI:10.1007/978-981-99-4783-2. View in Article Google Scholar
[7]	Yesilyurt H., Dokuz Y., Dokuz A.S., et al. (2024). Data-driven energy consumption prediction of a university office building. Energy 310:133242. DOI:10.1016/j.energy.2024.133242 View in Article CrossRef Google Scholar
[8]	Hsu P.C., Gao L., Hwang Y., et al. (2025). Comparative study of LSTM and ANN models for power consumption prediction of variable refrigerant flow (VRF) systems. Int. J. Refrig. 169:55−68. DOI:10.1016/j.ijrefrig.2025.02.014 View in Article CrossRef Google Scholar
[9]	Oh K. and Kim E.J. (2024). Predicting the energy consumption of a VRF heat pump using manufacturer performance data. Energy Build. 303:113798. DOI:10.1016/j.enbuild.2024.113798 View in Article CrossRef Google Scholar
[10]	Wang R., Bai L., Rayhana R., et al. (2024). Personalized federated learning for buildings energy consumption forecasting. Energy Build. 323:114762. DOI:10.1016/j.enbuild.2024.114762 View in Article CrossRef Google Scholar
[11]	Liu H., Liu Y., Huang H., et al. (2024). Energy consumption dynamic prediction for HVAC systems based on feature clustering deconstruction. Build. Simul. 17:1439−1460. DOI:10.1007/s12273-024-00892-7 View in Article CrossRef Google Scholar
[12]	Ali M., Kumar A. and Choi B.J. (2025). Privacy preserving federated learning for energy disaggregation of smart homes. IET Cyber-Phys. Syst.: Theory Appl. 10. DOI:10.1049/cps2.70013. View in Article Google Scholar
[13]	Mulik A., Mehta D., Mirgh R., et al. (2023). A federated learning approach to predict energy consumption. ICCCIS:187–192. DOI:10.1109/ICCCIS60361.2023.10344237. View in Article Google Scholar
[14]	Venkataramanan V., Kaza S., Annaswamy A.M., et al. (2023). DER forecast using privacy-preserving federated learning. IEEE Internet Things J. 10:2046−2055. DOI:10.1109/JIOT.2022.32280 View in Article CrossRef Google Scholar
[15]	Khalil M., Esseghir M., Merghem-Boulahia L., et al. (2021). Federated learning for energy-efficient thermal comfort service. IEEE GLOBECOM:1–6. DOI:10.1109/GLOBECOM46548.2021.9685634. View in Article Google Scholar
[16]	Zhou X., Wang N., Zou J., et al. (2024). Analysis and prediction of energy consumption in office buildings with variable refrigerant flow systems. J. Build. Eng. 97:110936. DOI:10.1016/j.jobe.2024.110936 View in Article CrossRef Google Scholar
[17]	Pachano J.E., Nuevo-Gallardo C. and Fernández Bandera C. (2025). An empirical comparison of a calibrated white-box versus multiple LSTM black-box building energy models. Energy Build. 333:115485. DOI:10.1016/j.enbuild.2025.115485 View in Article CrossRef Google Scholar
[18]	Yue B., Wei Z., Zheng C., et al. (2023). Power consumption prediction of variable refrigerant flow system through data-physics hybrid approach. Energy 278:127826. DOI:10.1016/j.energy.2023.127826 View in Article CrossRef Google Scholar
[19]	Liu Q., Yan Y., Jin Y., et al. (2024). Secure federated evolutionary optimization-a survey. Engineering 34:23−42. DOI:10.1016/j.eng.2024.02.004 View in Article CrossRef Google Scholar
[20]	Li A., Xiao F., Fan C., et al. (2021). Development of an ANN-based building energy model using transfer learning. Build. Simul. 14:89−101. DOI:10.1007/s12273-020-00677-9 View in Article CrossRef Google Scholar
[21]	Fan C., Lei Y., Sun Y., et al. (2022). Data-centric or algorithm-centric: Exploiting the performance of transfer learning for improving building energy predictions in data-scarce context. Energy 240:122775. DOI:10.1016/j.energy.2021.122775 View in Article CrossRef Google Scholar
[22]	Mustaffa Z., Sulaiman M.H. (2025). Advanced forecasting of building energy loads with XGBoost and metaheuristic algorithms. Energy Storage Sav. DOI:10.1016/j.enss.2025.1000305. View in Article Google Scholar
[23]	McMahan H.B., Moore E., Ramage D., et al. (2016). Communication-efficient learning of deep networks from decentralized data. arXiv:1602.05629. DOI:10.48550/arXiv.1602.05629. View in Article Google Scholar
[24]	Gao J., Wang W., Liu Z., et al. (2021). Decentralized federated learning framework for the neighborhood: a case study on residential building load forecasting. 19th ACM Conf. Embed. Net. Sens. Sys.:453–459. DOI:10.1145/3460312.3482513. View in Article Google Scholar
[25]	Wang Z., Yu P. and Zhang H. (2023). Privacy-preserving regulation capacity evaluation for HVAC systems. IEEE Trans. Smart Grid 14:3535−3549. DOI:10.1109/TSG.2023.32422 View in Article CrossRef Google Scholar
[26]	Lu Y., Cui L., Wang Y., et al. (2023). Residential energy consumption forecasting based on federated reinforcement learning. Comput. Model. Eng. Sci. 137:717−732. DOI:10.32604/cmes.2023.02876 View in Article CrossRef Google Scholar
[27]	Tang L., Xie H., Wang X., et al. (2023). Privacy-preserving knowledge sharing for few-shot building energy prediction. Appl. Energy 337:120860. DOI:10.1016/j.apenergy.2023.120860 View in Article CrossRef Google Scholar
[28]	Chaudhary R.K., Kumar R. and Saxena N. (2025). Ameliorating clustered federated learning using real-coded genetic algorithm. Cluster Comput. 28:415. DOI:10.1007/s10586-024-04867-1 View in Article CrossRef Google Scholar
[29]	Wu J., Ji H., Yi J., et al. (2025). Optimizing client selection in federated learning base on genetic algorithm. Cluster Comput. 28:400. DOI:10.1007/s10586-024-04869-z View in Article CrossRef Google Scholar
[30]	Zhai R., Chen X., Pei L., et al. (2023). A federated learning framework against data poisoning attacks. Electronics 12:560. DOI:10.3390/electronics12030560 View in Article CrossRef Google Scholar
[31]	Branco P., Torgo L., Ribeiro R.P. (2017). SMOGN: a pre-processing approach for imbalanced regression. Int. Workshop Learn. Imba. Dom. PMLR:36–50. DOI:10.48550/arXiv.1704.03466. View in Article Google Scholar
[32]	Rad M., Rafiei A., Grunwell J., et al. (2025). Tackling the small imbalanced horizontal dataset regressions by stability selection and SMOGN. Int. J. Med. Inform. 196:105809. DOI:10.1016/j.ijmedinf.2024.105809 View in Article CrossRef Google Scholar
[33]	Kocoglu Y., Gorell S.B., Emadi H., et al. (2024). Enhancing shale gas EUR predictions with TPE optimized SMOGN. Gas Sci. Eng. 131:205475. DOI:10.1016/j.gsxe.2024.205475 View in Article CrossRef Google Scholar
[34]	Meyes R., Lu M., Puiseau C.W., et al. (2019). Ablation studies in artificial neural networks. arXiv:1909.00608. DOI:10.48550/arXiv.1909.00608. View in Article Google Scholar
[35]	Zhao Y., Chen W., Xu Z., et al. (2025). AbGen: evaluating large language models in ablation study design. arXiv:2504.09641. DOI:10.48550/arXiv.2504.09641. View in Article Google Scholar
[36]	Kamalov F., Moussa S., Reyes J.A. (2023). Data transformation in machine learning. 2023 Int. Conf. Innov. Intel. Inform. Comput. Tech.:115–120. DOI:10.1109/3ICT58783.2023.10330167. View in Article Google Scholar
[37]	Ji J., Pang W., Zhou C., et al. (2012). A fuzzy k-prototype clustering algorithm for mixed numeric and categorical data. Knowl.-Based Syst. 30:129−135. DOI:10.1016/j.knosys.2012.01.015 View in Article CrossRef Google Scholar
[38]	Ji J., Bai T., Zhou C., et al. (2013). An improved k-prototypes clustering algorithm for mixed numeric and categorical data. Neurocomputing 120:590−596. DOI:10.1016/j.neucom.2013.04.011 View in Article CrossRef Google Scholar

Cited by

ABOUT THIS ARTICLE

Cite this article:

Guo Z., Wang B., Jiang Y., et al. (2026). Privacy-Preserving Federated Learning for VRF System Energy Use Prediction with SMOGN Augmentation And GA-Based Aggregation. Energy Use 2:100060. https://doi.org/10.59717/ipj.energy-use.2026.100060

Guo Z., Wang B., Jiang Y., et al. (2026). Privacy-Preserving Federated Learning for VRF System Energy Use Prediction with SMOGN Augmentation And GA-Based Aggregation. Energy Use 2:100060. https://doi.org/10.59717/ipj.energy-use.2026.100060

Welcome!

To request copyright permission to republish or share portions of our works, please visit Copyright Clearance Center's (CCC) Marketplace website at marketplace.copyright.com.

PDF

Figures(8)

Supplementary Information

download_supplementary
EU-2026-0028-SI

Article Metrics

Article views(241) PDF downloads(87)

Privacy-Preserving Federated Learning for VRF System Energy Use Prediction with SMOGN Augmentation And GA-Based Aggregation

GRAPHICAL ABSTRACT

PUBLIC SUMMARY

ABSTRACT