基于PCA-LVQ多联机系统制冷剂充注量故障诊断研究

Fault diagnosis of refrigerant charge based on the combination of principal component analysis and learning vectorized neural network

对于多联机空调系统,当发生了制冷剂的泄漏故障时,设备的性能会下降甚至失效,不仅造成能源的浪费、热舒适性的降低、设备有效使用寿命的缩短,还可能对环境造成严重污染。本论文将使用主成分分析(Principal Component Analysis, PCA)与学习向量化(Learning Vectorized Quantization Neural network, LVQ)神经网络结合的数据挖掘技术方法对多联机空调系统的制冷剂充注量故障进行诊断。首先通过多联机组试验获得运行数据,接着进行数据清洗工作,然后导入到PCA模型当中,利用权值矩阵将原始变量转换成新的综合变量,并将综合变量中重要度靠前的9个变量导入到LVQ,对多联机制冷剂充注量的故障进行诊断。结果表明:主成分分析与学习向量化神经网络的联合诊断模型结构简单、训练速度快、易于实现(代码简单),诊断正确率高,相较于单纯的LVQ算法模型,对制冷剂充注量故障的诊断效果更优。

For a multi-line air conditioning system,when a refrigerant leakage failure occurs,the performance of the equipment will decrease or even fail,which not only causes a waste of energy,a decrease in thermal comfort,and a shortening of the effective service life of the equipment,but also may affect the environment.Cause serious pollution.This paper will use principal component analysis(Principal Component Analysis,PCA)and learning vectorization(Learning Vectorized Quantization Neural network,LVQ)neural network combined data mining technology method for multi-line air conditioning Diagnose the system’s refrigerant charge failure.First obtain the operating data through multi-online group experiments,and then perform data cleaning,import it into the PCA model,use the weight matrix to convert the original variables into new comprehensive variables,and change the top 9 variables of the comprehensive variables.Import to LVQ to diagnose the fault of multi-line refrigerant charge.the result surface:the joint diagnosis model of principal component analysis and learning vectorized neural network is simple in structure,fast in training,easy to implement(simple code),and high in diagnosis accuracy.Compared with the pure LVQ algorithm model,the refrigerant charge amount,the fault diagnosis effect is better.