The invention discloses a power system fault diagnosis method based on time sequence similarity matching. The method comprises the steps of searching for a passive area in a system, determining a faulty region, and obtaining a suspicious fault component set; forming a warning hypothesis time sequence set based on a protective equipment configuration model; calculating the distance between a sub-time sequence hypothesis and a warning information time sequence received by an SCADA system, obtaining the confidence of the time sequence hypothesis and the confidence of component failures, and identifying fault components; conducting faulty region identification after an accident is over usually, that is when a dispatching center has already receives relevant complete warning information, conducting fault diagnosis by means of the matching method of static time sequences in the warning information, and then evaluating the behaviors of protection and circuit breakers. According to the method, by means of the time-order characteristics of the warning information sequence, situations such as failed warning and misinformation can still be identified quickly under the conditions of complicated failures, cascading failures and the like, the failure components and failure type can be diagnosed correctly, and a relay protection device can be evaluated.