基于学习向量量化神经网络的多联机系统制冷剂充注量故障诊断研究

Research on Fault Diagnosis of Refrigerant Charge in Variable Refrigerant Flow Air-conditioning System Based on Learning Vector Quantization Neural Network

针对暖通空调系统故障而导致的建筑能耗增长问题,本文提出了一种基于学习向量量化(LearningVector Quantification,LVQ)神经网络的制冷剂充注量故障诊断模型,故障诊断分为数据预处理、建立初始模型、LVQ 模型的训练和仿真测试 4 个步骤,并对隐含层节点数进行了参数寻优。实验共设置 9 种制冷剂充注量水平,经过数据预处理后选取了 12 个特征变量,建立了 LVQ 神经网络建模。将经过数据预处理后的数据集以 75%∶25%的比例划分为训练集和测试集,分别用于研究训练和测试模型性能。结果表明:在制冷剂充注量 LVQ 模型故障诊断中,制冷剂充注量适中、过量和不足的正确率分别为 52.5%、70.1%和87.5%,总体故障诊断正确率达到 70.0%。

Aiming at the problem of building energy consumption growth caused by the fault of HVAC system, a fault diagnosis model for refrigerant charge based on Learning Vector Quantification (LVQ) neural network is proposed in this paper. The fault diagnosis is divided into four steps: data preprocessing, initial model establishment, LVQ model training and simulation test, and the parameters of the model is optimized. After data preprocessing, 12 feature variables are selected to establish LVQ neural network model with 9 refrigerant charge levels. The pre-processed data are divided into a training set and a test set by a ratio of 75%/25%, which are used to study the performance of training and test models respectively. The results show that the accuracy of moderate, excessive and insufficient refrigerant charge in the LVQ model fault diagnosis are 52.5%, 70.1% and 87.5%, respectively, and the overall fault diagnosis accuracy rate is 70.0%.