The invention discloses a power grid fault diagnosis and alarm method based on multi-source data fusion. The method comprises the steps that a fault event is defined; delay intervals of main protection, near backup protection and far backup protection relative to the event C and delay time of protection and a breaker action are defined; relevant time reference points are acquired according to different independent data sources, and a fault element set under a time sequence constraint condition is obtained; the probability of occurrence of a fault of each fault element is acquired, and a fault confidence set of power elements is obtained through combination and subjected to normalization processing; a fast wavelet transformation and wavelet energy analysis method based on multiple resolutions is adopted to analyze discrete signals y(n); electrical quantity data is acquired, fault differential wave signals are acquired, voltage differential wave signals are subjected to wavelet energyanalysis, and corresponding values of energy distortion degrees are obtained; the energy distortion degrees of the fault elements are subjected to normalization processing, and a fault probability set of the power elements is obtained; and based on information fusion in a D-S evidence theory, fault confidence values of the fault elements are obtained.