The invention discloses a non-intrusive power load identification method based on a stack type auto-encoder. The method comprises the following steps of constructing a load characteristic data set ofpower load, determining the number S of initial input data sample points, and carrying out dimension reduction processing; establishing a stack type auto-encoder deep learning model to obtain featureinformation of the original power load current; and finally, processing the obtained feature information of the original power load current by adopting a Softmax function in machine learning deep learning. The Softmax is used as a classifier to classify and identify power utilization behaviors. The method solves the problems that in the prior art, when multiple loads work at the same time, the identification effect is poor, and the identification speed is low, solves the problems that in a scene, when the active power and the reactive power of electric appliances are close, the identificationeffect is poor, and improves the accuracy and the decomposition efficiency of power load decomposition.