The invention provides a turnout action reference curve selection method and an application thereof. The method comprises the steps of obtaining N action curves of a same turnout through a turnout microcomputer monitoring system, obtaining action time of the N curves, selecting the action time with a highest repetition number in the N action time, and marking the action time with the highest repetition number as Tx; assuming that M curves with action time lengths of Tx exist, and renumbering the M curves; for the M curves, dividing each curve into P equally divided parts according to the action time, and for a current value of a jth moment of the ith curve, obtaining current value matrixes I of the M curves in sequence for all the M curves; for M current values of any moment tj, calculating out a clustering center by utilizing a clustering algorithm, and marking a current value of the clustering center as C(j), wherein C(j) is the current value of the clustering center; and calculatingclustering centers of P moments in sequence, and by analyzing the clustering centers, selecting out a curve with a highest clustering frequency as a reference curve. Based on the reference curve, anabnormal curve diagnosis method is obtained by utilizing a similarity method.