基于Mt-SNE的冷水机组故障特征检测研究

FAULT DIAGNOSIS OF CHILLERS BASED ON MT-SNE

针对欧氏距离在高维空间中不能忠实反映样本位于非线性流形上的相似关系,提出基于马氏距离的t分布随机邻域嵌入算法(Mt-SNE)。用改进的算法对输入数据进行降维处理,并将输出结果作为支持向量机(SVM)的输入向量进行分类,采用ASHRAE制冷系统故障实验数据进行模型训练与验证。结果表明,该算法比传统的线性降维算法主元分析法(PCA)和非线性降维算法t分布随机邻域嵌入(t-SNE)都有更好的特征提取能力,可以用于冷水机组故障数据的特征提取。

Aiming at the fact that the Euclidean distance cannot truly reflect the similarity of samples located on nonlinear manifolds in high-dimensional space, this paper proposes a t-distribution random neighborhood embedding algorithm based on Mahalanobis distance(Mt-SNE). The improved algorithm was used to reduce the dimensionality of the input data, and the output result was taken as the input vector of the support vector machine(SVM) for classification. The ASHRAE refrigeration system fault experimental data was applied for model training and verification. The results show that the Mt-SNE algorithm has better feature extraction capabilities than traditional linear dimensionality reduction algorithm principal component analysis(PCA) and nonlinear dimensionality reduction algorithm t-distributed random neighborhood embedding(t-SNE), which can be used for feature extraction of chillers fault data.