当前位置: X-MOL 学术Energy Build. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Study on enhanced fault diagnosis of chiller units in HVAC systems under the imbalanced data environment using GA-Optimized LightGBM
Energy and Buildings ( IF 6.6 ) Pub Date : 2025-01-22 , DOI: 10.1016/j.enbuild.2025.115360
Manfeng Li, Haoyuan Tao, Mengke Liu, Tianbiao He

HVAC systems consume about 50% of the total energy in commercial buildings, with chiller units playing a central role but being prone to complex failures. These undetected failures of the chiller units lead to energy waste, equipment damage, and potential safety risks, emphasizing the critical need for early and accurate fault detection. This study proposes a data-driven fault diagnosis framework for chiller units that addresses feature selection, class imbalance, and model optimization to improve diagnostic accuracy. The Relief-Family algorithm ranks feature importance, guiding the development of fault diagnosis models, including Random Forest, Support Vector Machine, BP Neural Network, and LightGBM. To tackle data imbalance between normal and fault samples, the SMOTEENN-CGAN approach synthesizes minority fault data, enabling the creation of the SC-LightGBM model, which effectively improves fault detection across different severity levels. The model’s hyperparameters are further refined using genetic algorithm (GA) and particle swarm optimization (PSO) to maximize performance. Comparative analysis shows that the GA-LightGBM model outperforms other methods, with accuracy improvements of 1.16%, 1.64%, and 3.36% under fault-to-normal data ratios of 10:1, 20:1, and 50:1, respectively. These findings demonstrate that the GA-LightGBM method offers a robust and reliable solution for fault diagnosis in chiller units, contributing to enhanced efficiency and safety in HVAC systems.

中文翻译:


基于GA优化LightGBM的不平衡数据环境下暖通空调系统冷水机组增强故障诊断研究



HVAC 系统消耗了商业建筑总能源的 50% 左右,冷水机组起着核心作用,但容易出现复杂的故障。这些未检测到的冷水机组故障会导致能源浪费、设备损坏和潜在的安全风险,因此迫切需要及早准确地检测故障。本研究提出了一个数据驱动的冷水机组故障诊断框架,该框架解决了特征选择、类别不平衡和模型优化问题,以提高诊断准确性。Relief-Family 算法对特征重要性进行排名,指导故障诊断模型的开发,包括随机森林、支持向量机、BP 神经网络和 LightGBM。为了解决正常样本和故障样本之间的数据不平衡问题,SMOTEENN-CGAN 方法合成了少数故障数据,从而能够创建 SC-LightGBM 模型,从而有效地改进了不同严重性级别的故障检测。使用遗传算法 (GA) 和粒子群优化 (PSO) 进一步优化模型的超参数,以最大限度地提高性能。对比分析表明,GA-LightGBM 模型优于其他方法,在 10:1、20:1 和 50:1 的故障与正常数据比率下,准确率分别提高了 1.16%、1.64% 和 3.36%。这些发现表明,GA-LightGBM 方法为冷水机组的故障诊断提供了一种稳健可靠的解决方案,有助于提高 HVAC 系统的效率和安全性。
更新日期:2025-01-22
down
wechat
bug