改进LNS和邻域保持嵌入算法的研究

MODIFIED LNS AND NEIGHBORHOOD PRESERVING EMBEDDING ALGORITHM

传统邻域保持嵌入算法(Neighbor Preserving Embedding,NPE)对具有多中心、方差差异明显特性的高维数据的降维处理效果并不好,因此提出一种改进LNS和邻域保持嵌入算法(Modified Local Neighbor Standardization-Neighbor Preserving Embedding,MLNS-NPE),并应用于故障诊断中。利用MLNS算法对数据进行处理,对处理后的数据进行NPE算法建模。在数值例子和青霉素发酵过程中应用该算法与传统NPE算法、核邻域保持嵌入算法(KNPE)、KNN算法比较,结果表明,采用该算法后,数据多中心和模态差异消除,为后续NPE算法的应用提供先决条件,同时相比其他算法故障检测率最高,提高了NPE算法对多模态数据的检测能力。

Traditional neighbor preserving embedding(NPE)is not effective in dimensionality reduction of high-dimensional data with multi-center and obvious variance differences.Therefore,an improved LNS and neighbor preserving embedding(MLNS-NPE)algorithm is proposed and applied to fault diagnosis.It used MLNS algorithm to process the data,then carried out NPE algorithm modeling on the processed data.In the process of numerical examples and penicillin fermentation,the comparison results between the algorithm and traditional NPE algorithm,kernel neighborhood preserving embedding algorithm(KNPE)and KNN algorithm show that after the algorithm is adopted,the data multicenter and modal difference are eliminated,which provides a prerequisite for the application of subsequent NPE algorithm.At the same time,compared with other algorithms,the fault detection rate is the highest,and the detection capability of NPE algorithm for multimodal data is improved.