| 摘要: |
| 针对冷水机组中常见的7种故障，本文基于交叉熵损失函数和随机梯度下降算法建立了多目标异权重回归模型，进行故障诊断。该模型较常规的机器学习分类模型简单，无需迭代，计算速度快，且为显式模型（非黑箱），可直观分析各参数对每类故障的重要程度。与传统的单目标回归模型相比，故障诊断性能优势显著，在不同特征集合下，性能最低提升40.50%。对比不同文献中特征集合在本模型中的效果，并提出了新的特征集合，正常运行及7类故障的总体诊断准确率可达89.83%，局部故障的诊断准确率达到98%以上。通过可视化诊断模型中的参数权重，发现过冷度和供油温度参数对诊断制冷剂泄漏、制冷剂过充和润滑油过量3种系统性故障最为重要；供油压力、冷凝器趋近温度、蒸发器与冷凝器的水流量参数对诊断4种局部故障最为重要。 |
| 关键词: 冷水机组 故障诊断 显式模型 交叉熵 随机梯度下降 |
| DOI： |
| 投稿时间：2022-11-19 修订日期：2023-02-24
录用日期：2023-04-18 |
| 基金项目:国家自然科学基金（51506125）资助项目。 |
|
| Explicit Model for Chiller Fault Diagnosis Based on Multi-objective Regression with Different Weights |
|
Wu Kongrui1, Han Hua1, Yang Yuting1, Lu Hailong1,2, Ling Minbin1
|
| (1.School of Energy and Power Engineering, University of Shanghai for Science and Technology;2.CQ Midea General Refrigeration Equipment Co.,Ltd.) |
| Abstract: |
| Based on the cross-entropy loss function and stochastic gradient descent algorithm, a weight regression fault diagnosis model was established for seven common faults in a chiller. The weighted regression model was slightly more complex than the pure linear regression model; however, the fault diagnosis performance was clearly better, and the minimum performance was improved by 40.50% under different feature sets. When comparing the effects of feature sets from various sources in this model and introducing a new feature set, the accuracy reached 89.83%. Notably, the diagnostic accuracy for local faults exceeded 98%. The explicit model for chiller fault diagnosis is summarized, and by examining the parameter weights in the visual diagnosis model, it was determined that the oil supply pressure, oil supply temperature, and degree of subcooling were the most crucial parameters for diagnosing three types of system faults. Conversely, the refrigerant pressure in the condenser, temperature difference in the condenser, and water flow parameters between the evaporator and condenser were identified as the most important parameters for diagnosing four local faults. |
| Key words: chiller fault diagnosis explicit model cross entropy stochastic gradient descent |
