| 11 | 0 | 18 |
| 下载次数 | 被引频次 | 阅读次数 |
提出了一种基于卷积神经网络（CNN）-微调（FT）的故障诊断迁移方法，利用信息丰富的源域多联机系统的先验知识来建立目标多联机系统的诊断模型。首先对源域进行预训练，通过参数寻优找到最优CNN模型；然后将预训练模型迁移至目标域上，只用少量目标数据训练CNN顶层，准确率为86.71%；依次解冻前面的网络层并进行微调处理，准确率升至95.83%，显著优于目标域特定训练（81.02%）、源域模型直接迁移（33.45%）2种情况。
Abstract：A fault diagnosis migration method based on CNN-FT(convolutional neural network with fine-tuning) is proposed, which leverages the informative prior knowledge from the source-domain variable refrigerant flow system to establish a diagnostic model for the target variable refrigerant flow system. Firstly, the source-domain is pre-trained, and the optimal CNN model is found by parameter optimization. Then the pre-training model is migrated to the target domain, and only a small amount of target data is used to train the top layer of CNN, with an accuracy of 86.71%. The previous network layer is thawed in turn for fine-tuning, and the accuracy rate is improved to 95.83%, which is significantly better than the target domain specific training(81.02%) and the source domain model direct migration(33.45%).
[1] WAN H, CAO T, HWANG Y, et al. A review of recent advancements of variable refrigerant flow air-conditioning systems[J]. Applied Thermal Engineering, 2020, 169:114893.
[2]机电信息编辑部. 2020年1月-9月中国中央空调市场总结报告[J].机电信息, 2020(31):18-24.
[3] ZHANG G, XIAO H, ZHANG P, et al. Review on recent developments of variable refrigerant flow systems since2015[J]. Energy and Buildings, 2019, 198:444-466.
[4]袁玥,陈焕新,石书彪,等.基于主成分分析和神经网络相结合的制冷剂充注量故障诊断[J].制冷技术, 2017,37(6):45-50.
[5]范波,丁云霄,纪轲,等.基于类随机森林算法和仿真的多联机故障诊断方法研究[J].制冷技术, 2021, 41(6):21-28.
[6]王誉舟,李正飞,魏文天,等.基于递归特征消除-加权k近邻算法的多联机系统制冷剂充注量故障诊断策略[J].制冷技术, 2020, 40(1):16-22.
[7]庄福振,罗平,何清,等.迁移学习研究进展[J].软件学报, 2015, 26(1):26-39.
[8] CAO P, ZHANG S, TANG J. Preprocessing-free gear fault diagnosis using small datasets with deep convolutional neural network-based transfer learning[J]. IEEE Access,2018, 6:26241-26253.
[9] YU X, CHEN W, WU C, et al. Rolling bearing fault diagnosis based on domain adaptation and preferred feature selection under variable working conditions[J].Shock and Vibration, 2021, 99:1-27.
[10] LIU J, ZHANG Q, LI X, et al. Transfer learning-based strategies for fault diagnosis in building energy systems[J].Energy and Buildings, 2021, 250:111256.
[11] ZHU X, CHEN K, ANDUV B, et al. Transfer learning based methodology for migration and application of fault detection and diagnosis between building chillers for improving energy efficiency[J]. Building and Environment,2021, 200:107957.
[12] FAN C, SUN Y, XIAO F, et al. Statistical investigations of transfer learning-based methodology for short-term building energy predictions[J]. Applied Energy, 2020, 262:114499.
[13] MA J, CHENG J C P, JIANG F, et al. A bi-directional missing data imputation scheme based on LSTM and transfer learning for building energy data[J]. Energy and Buildings, 2020, 216:109-114.
[14]陶昊园,范波,李满峰.基于改进神经网络的冷水机组故障诊断研究[J].制冷技术, 2023, 43(2):43-51.
[15]刘卓轩,曹承,张天乐,等.基于共形预测的冷水机组鲁棒故障诊断方法研究[J].制冷技术, 2024, 44(4):7-15.
[16] LU W, LIANG B, CHENG Y, et al. Deep model based domain adaptation for fault diagnosis[J]. IEEE Transactions on Industrial Electronics, 2016, 64(3):2296-2305.
[17]许源驿,赵思哲,陈焕新,等.基于生成对抗网络学习的建筑暖通空调负荷特征能耗研究[J].制冷技术, 2024,44(6):50-58.
[18] FANG X, GONG G, LI G, et al. A hybrid deep transfer learning strategy for short term cross-building energy prediction[J]. Energy, 2021, 215:1-6.
[19]周飞燕,金林鹏,董军.卷积神经网络研究综述[J].计算机学报, 2017, 40(6):1229-1251.
[20] XIE Y, SUN W, REN M, et al. Stacking ensemble learning models for daily runoff prediction using 1D and 2D CNNs[J]. Expert Systems with Applications, 2023, 217:119469.
[21] XU H, LI W, CAI Z. Analysis on methods to effectively improve transfer learning performance[J]. Theoretical Computer Science, 2023, 940:90-107.
[22] YUAN F, ZHANG Z, FANG Z. An effective CNN and Transformer complementary network for medical image segmentation[J]. Pattern Recognition, 2023, 136:109228.
[23]?ELIKKANAT A, MALLIAROS F D. Topic-aware latent models for representation learning on networks[J]. Pattern Recognition Letters, 2021, 144:89-96.
[24] QIAO J, WANG G, LI W, et al. An adaptive deep Qlearning strategy for handwritten digit recognition[J].Neural Networks, 2018, 107:61-71.
基本信息:
DOI：
中图分类号:TP183;TU831.3
引用信息:
[1]蒋敏辉,陈焕新,苟伟.基于卷积神经网络-微调的多联机故障诊断迁移研究[J].制冷技术,2025,45(02):22-29+49.
基金信息:
国家自然科学基金（No.51876070）