针对工业锅炉的动态特性与多模态特性带来的故障检测问题，提出一种动态加权差分主成分分析法(DWDPCA)。首先，建立合理的时间窗描述系统的时序特性；然后，对时间窗中的样本寻找其空间上的第一近邻和第一近邻的近邻集，使用加权差分方法将数据转化为单模态结构；最后，利用处理后的数据建立PCA模型进行故障检测。通过在某实际工业锅炉中的应用表明，DWDPCA方法可解决动态时序问题和多模态数据中心漂移问题，显著提高故障检测的精度。

A novel dynamic weighted differential principal component analysis (DWDPCA) was proposed to address the challenges of detecting faults in industrial boilers with dynamic and multimodal characteristics. Firstly, a reasonable time window was established to capture the time sequence characteristics of the system; then, the first nearest neighbor and its neighbor set in space were searched for the samples within the time window. The data was transformed into a singlemodal structure using a weighted differential method; finally, the processed data established a PCA model for accurate fault detection. The application of the DWDPCA method in an industrial boiler, which has shown promising results in solving the dynamic time sequence and center drift problems associated with multimodal data, has significantly improved fault detection accuracy.

TP273

国家自然科学基金(52071047，62073054)