基于稀疏自编码器-支持向量机的空调制冷系统故障诊断

Fault Diagnosis of Air-conditioning Refrigeration System Based on Sparse Auto Encoder-support Vector Machine

针对目前空调制冷系统内部的多参数性与物质复杂性等现状,本文提出了一种基于稀疏自编码器-支持向量机的空调制冷系统故障诊断方法,研究了其在空调制冷系统故障诊断领域的应用潜力。故障诊断结果表明,单层稀疏自编码器模型的故障诊断正确率为95.47%。采用稀疏自编码器特征提取效果优于主元分析法,隐藏层层数和节点数对稀疏自编码器诊断性能有较大的影响,多层稀疏自编码器性能优于单层稀疏自编码器。本系统中隐藏层层数为4,节点数分别为600、500、400和300的多层稀疏自编码器模型诊断性能最优。

Aiming at the current multi-parameter and material complexity of air-conditioning refrigeration systems,a fault diagnosis method for air-conditioning refrigeration system based on sparse auto encoder-support vector machine was proposed,and its application potential in air-conditioning refrigeration system was studied.The results show that the fault diagnosis accuracy of the single-layer sparse auto encoder model is 95.47%.In addition,the feature extraction effect of sparse auto encoder is better than the principal component analysis method.The number of hidden layers and nodes have great influence on the diagnostic performance of sparse auto encoder.The performance of multi-layer sparse auto encoder is better than single-layer sparse auto encoder.In this system,the multi-layer sparse self-encoder model with 4 hidden layers and 600,500,400 and 300 nodes has the best diagnostic performance.