Nonnegative Matrix Factorization And Its Application In The Anomaly Detection Of Hydroelectric Generating Unit

With the development of science and technology, national pay more and more attention to the development of renewable energy. Small hydropower as a clean, sustainable, safe and effective renewable energy, is of far-reaching significance in the state energy development strategy. Since the equipment of hydroelectric generating unit is more complex, we usually adopt the artificial way to detect the abnormal fault. But this way is inefficient and sometimes can’t judgment the accurate position of the noise source. So that it is necessary to consider the machine learning method. Because nonnegative matrix factorization has the property of nonnegative combination after decomposition, the research of applying the nonnegative matrix factorization algorithm to the hydroelectric generating unit noise source detection is valuable.This paper analyses the characteristics of the hydroelectric generating unit noise signal and the problems when applying the nonnegative matrix factorization algorithm to the hydroelectric generating unit noise source detection, and then designs the corresponding nonnegative matrix factorization algorithm. The main contributions of the work are as follow:1) Design an incremental nonnegative matrix factorization algorithm based on BSS model, INMF. When dealing with huge amounts of data, the existing NMF algorithm need to recalculate all the data at a time, greatly reduces the efficiency. This algorithm not only has the same dimension reduction effect but also has a good ability of incremental calculation compared with the original algorithm. Finally the experiment on a public database also verifies the validity of the algorithm.2) Design an incremental locality preserving nonnegative matrix factorization method based on Euclidean distance, ILPNMF. Since the abnormal vibration noise is hard to get in the early stage of the noise detection, the online learning is needed. We combine NMF with LPP and then introducing the idea of increment learning, to make it more applicable to complex industrial equipment such as hydroelectric generating unit. ILPNMF can decrease the time and space of matrix decomposition during the process of online learning.3) At the end the paper analyses the reason characteristics of abnormal vibration of the hydroelectric generating unit, then applies ILPNMF to hydroelectric generating unit noise source online detection. The simulation results show that ILPNMF on hydroelectric generating unit noise source detection has high recognition rate and efficiency.