The invention discloses a transformer substation breaker secondary loop fault prediction method considering correlation coefficients, comprising the following steps: 1) outputting an installation manufacturer, an operating period, a maintenance frequency, an operating environment, a type of power distribution device, and a load situation as historical data from a PMS accounting system, preprocessing and normalizing a part of the data as training samples; 2) calculating correlation coefficients for the training samples and samples to be evaluated and setting a threshold, removing the training samples below the threshold and selecting samples with a higher relevance to the samples to be evaluated; 3) determining optimal parameters of a support vector machine by using a Gridregression searchfor the selected training samples; 3) using a function train-svm in a LIBSVM toolbox to calculate, obtaining a deviation b and Lagrangian coefficients alpha, [alpha]*, and obtaining a prediction model: (img file='DDA0001752017490000011. TIF' wi='671' he='122'/); 4) using the prediction model to predict the samples to be evaluated; if an output is -1, the breaker is operating normally; if the output is +1, the breaker is faulty.