Stochastic Neighbor Embedding Analysis Method And Its Application In Fault Diagno- SIS Of Hydroelectric Generating Sets

For being main equipment, hydroelectric generating set’s health is not only related with the security of hydro-power plant, but also with the fulfillment of providing safe and stable electricity. Many unfavorable factors are increasingly impacting the security and stability of power grid, such as:the complexities of the equipment, seasonal variations of the units and a variety of the abnormal vibrations etc. So, it is of great significance to monitor the motor running and diagnose faults.The conventional ways of fault diagnosis, which still play the leading role in the hydropower units, is based on the experience and knowledge of professional and technical personnel. Its disadvantages become evident, so it has to be raised the degree of equipment automation and intelligence.This paper analyzes characteristics of stochastic neighbor embedding analysis series, and applies it to fault diagnosis of hydroelectric generating sets. The main contributions of the work are as follow:(1) A novel linear supervised feature extraction method named discriminative stochastic neighbor embedding is proposed based on the algorithm of stochastic neighbor embedding that is unsupervised and nonlinear. DSNE selects the joint probability to model the pairwise similarities of input samples with class labels. Then, it uses the linear projection matrix to discover the underlying structure of data manifold which is nonlinear. Experimental results suggest that DSNE provides a better visualization effectiveness as well as powerful pattern revealing capability for complex manifold data.(2) A fast discriminative stochastic neighbor embedding analysis is proposed by improving the existing DSNE. FDSNE adopts an alternative probability distribution model constructed based on its K nearest neighbors from the inter-class and intra-class samples. Experimental results on several datasets show that FDSNE not only enhances the computational efficiency but also obtains higher classification accuracy.(3) A kernel-based discriminative stochastic neighbor embedding analysis method is proposed by imposing the kernel trick, which furthest maintains the observation information and effectively improves the performance of dimensionality reduction. Based on DSNE, the proposed method introduces kernel function and maps the data into a high-dimensional feature space, and then it selects the joint probability to model the pairwise similarities of input samples. KDSNE outstands the feature differences between inter-class samples and makes the samples linear separable so as to improve the classification performance.(4) KDSNE is applied to fault diagnosis of hydroelectric generating sets by shaft center orbit feature extraction. Experiments demonstrate that KDSNE is effective and feasible.