The invention discloses a low-frequency oscillation on-site online identification method based on an extended Prony algorithm, which comprises the steps of computing an active power, a voltage and power factor product Ucosphi and a current zero-sequence component according to three-phase voltage current phasor data that are calculated by a synchronous phasor measuring device; determining whether low-frequency occurs by means of a Mandani fuzzy logic, and if yes, performing parameter identification by means of a Prony algorithm, computing for obtaining amplitude, frequency, damping factor and damping ratio of low-frequency oscillation, thereby realizing on-site online identification for the low-frequency oscillation, and obtaining high identification speed and high accuracy. Software and hardware expenditure at a WAMS main station end can be greatly reduced. Furthermore the low-frequency oscillation on-site online identification method supplies reliable original data and data support for the functions of full-system electric network wide-area monitoring, transformer station automatic measuring and controlling, stable controlling, adaptive relay protecting, etc.