The classification and identification of voltage sag sources has the problem of having incomplete information. Given the weak learning characteristics of the existing single identification method and the strong consistency of the combined identification method, a sag source identification method based on heterogeneous stacking ensemble learning is proposed to improve the generalization ability and robustness of the recognition model. Line faults are subdivided into common faults and lightning faults, and ten types of single voltage sag sources are used as the identification target. Nine waveform statistical parameters that can characterize the differences are selected to construct a 27-dimension recognition vector. The stacking ensemble algorithm is introduced, five highly differentiated single recognition methods are used as base-classifiers, and random forest is selected as the meta-classifier to establish a heterogeneous stacking ensemble recognition model. Through the verification of a PSCAD simulation model and measured data, and comparison with six typical sag source identification methods, it is shown that the proposed method has high identification accuracy and good anti-noise performance with good engineering value. This work is supported by the National Natural Science Foundation of China (No. 51807126).