The invention relates to an emission monitoring time series data abnormal value detection method for a coal-fired unit. Specifically, all-working-condition historical sample data set of a measured point is subjected to working condition division by applying a k-means clustering algorithm according to unit operation states and environment-protection facility operating ways; then specific to each working condition historical sample data set, an abnormal value detection model is established based on a statistical hypothesis test method; and the time series data of the measured point is subjected to online identification by selecting different hypothesis test methods according to data distribution characteristics of the working condition sample set, and then abnormal points which are inconsistent with historical distribution and current change tendency can be positioned in time. Abnormal value online detection on the field monitored data is performed by the hypothesis test methods, and the abnormal points which are inconsistent with historical distribution or current change tendency can be dynamically identified, so that early warning for more complex fault monitoring is provided, refine degree of alarming is improved, and online monitoring management level and rapid response capability are improved.