The invention discloses a data flow online abnormity detection method based on integrated learning, and relates to the technical field of data processing. According to the method, firstly, a Bagging integrated learning framework is applied, a stable LSTM prediction model is obtained through multiple times of iterative training of an LSTM model, and normal-complex scene data flow is achieved; the depth of the abnormal sample is identified; Meanwhile, a payload data flow is used as input; On one hand, real-time test data is provided for a stable LSTM model; secondly, a Bagging integrated learning framework is applied, a plurality of weak learners are integrated to obtain strong learners, a learning device based on a Stacking algorithm is established, an optimal detection result is obtained by combining output results of the weak learners, and the accuracy of data flow online abnormity detection is improved; And an abnormal detection result with better precision is obtained, and the falsealarm rate and the missing alarm rate are reduced. The problem that a traditional anomaly detection method cannot accurately mine the potential anomaly of the effective load in the complex space is solved.