The invention discloses an online power grid fault diagnosis method based on multisource data characteristic unit combination judgment. The method comprises the following steps that S1, single data sources are acquired from multisource data, and multiple characteristic units are split out of each data source according to the characteristics of the data sources; S2, the characteristic units of different data sources are combined so that multiple fault inference rules are formed, and the fault inference rules are stored in a rule base; and S3, operation data of equipment in the power grid operation process are acquired online, matching with the fault inference rules in the rule base is performed one by one, and alarm is performed if matching succeeds and judgment criteria are provided. A power grid fault can be timely and effectively diagnosed, and alarm can be timely and accurately generated so that the power grid real-time operation demand can be greatly met.