针对热工多模态过程的模式识别和聚类问题，提出了一种基于局部保留投影（Local Preserving Projection， LPP）子空间的混合聚类方案。首先，将高维的多模式过程数据通过局部保留投影方法投射到低维子空间中，在剔除噪声、提高计算效率的同时保留局部结构；其次，在LPP子空间中，结合传统的分层和非分层聚类算法的优点，使用凝聚kmeans算法，为多模式过程数据生成最佳的集成聚类解决方案。以某600 MW机组脱硫系统的多模式过程数据的识别与聚类过程为例，证明了该方法的有效性和实用性。

Considering the pattern recognition and clustering problems of thermal multimode processes, a hybrid clustering scheme based on the local preserving projection (LPP) subspace was proposed. Firstly, the data for the highdimensional multimode process is projected into the lowdimensional subspace through the LPP method, which eliminates the noise, improves computational efficiency and also retains the local structure. Secondly, combining the advantages of the traditional hierarchical and nonhierarchical clustering algorithms in the LPP subspace, the agglomerative kmeans algorithm is used to generate the best clustering solution for multimode process data. Taking the recognition and clustering processes of a certain 600 MW unit desulfurization system with the data of multimode processes as an example, the effectiveness and practicability of this method are proved.

TM621

中国博士后科学基金（2020M680474）