汽动给水泵是火电厂汽水系统重要的辅机设备，但给水泵运行环境恶劣，且故障种类多，导致给水泵故障频发，对电厂的经济性和安全性造成了影响。对此，提出了一种基于改进的层次凝聚聚类（Hierarchical Agglomerative Clustering，HAC）和多元状态估计（Multivariate State Estimation，MSET）的给水泵故障预警方法，首先选取与给水泵故障相关测点的历史数据，使这些数据可以涵盖给水泵正常运行时所有动态变化情况；通过主元分析法（Principal Component Analysis，PCA）和改进的小波去噪对数据进行预处理，实现数据的降噪和降维，再采用层次凝聚聚类算法构建记忆矩阵D，并引入了距离检测的方法对HAC进行改进，以此构建MSET预警模型，最后通过滑动窗口法分析残差，实现故障预警，并和最小二乘支持向量机（LSSVM）预警模型比较分析，经验证该模型可以准确高效地实现给水泵的早期预警。

The steamdriven feed water pump is an important auxiliary equipment for the steamwater system of the thermal power plant, but the operation environment of the feed water pump is harsh, and there are many types of failures, resulting in frequent failure of the feed water pump, which has an impact on the economy and safety of the power plant. Therefore, this paper proposes a feed water pump fault early warning method based on improved hierarchical agglomerative clustering (HAC) and multivariate state estimation technique (MSET). Firstly, the historical data of relative measuring point of feed water pump fault is selected to cover all dynamic changes in the normal operation of the feed water pump; through principal component analysis (PCA) and improved wavelet denoising to preprocess the data to achieve data noise reduction and dimensionality reduction; then, the memory matrix D is constructed by the HAC algorithm, and the distance detection method is introduced to improve the HAC, so as to construct the MSET early warning model; finally, the residual error is analyzed by the sliding window method to realize the fault early warning, and compared with the least squares support vector machine (LSSVM) early warning model. It is verified that the model can accurately and efficiently realize the early warning of the feed water pump.

TM621

上海市“科技创新行动计划”地方院校能力建设专项项目（19020500700）