The invention relates to a fault diagnosis method based on multi-feature information weighted fusion under spectral clustering analysis. The method comprises the steps of firstly, carrying out spectral clustering analysis on fault equipment; secondly, obtaining the reliability of a local diagnosis evidence of each SVM to each fault mode; thirdly, constructing basic probability distribution througha local diagnosis hard output judgment matrix of each SVM; fourthly, performing weighted processing on the basic probability distribution; fifthly, obtaining the credibility and the uncertainty; andfinally, through a set diagnosis rule, and in combination with the credibility and the uncertainty, performing diagnosis. Compared with the prior art, the method has the advantages that the situationthat evidences of different sources have different reliability for identification of propositions in an identification framework is considered, the conflict between the local diagnosis of the SVMs isreduced, effective combination of the SVMs and an improved evidence theory is realized, and the shortcoming that a synthetic result cannot reflect an objective fact due to the unreliability of identification is overcome.