The invention relates to a method for establishing an abnormal power consumption detection model. The method comprises the following steps: extracting effective features of power consumption information of a training set user, and establishing an initial training data set for depth learning; in the initial training data set, using data without abnormal power consumption mark as the input of the abnormal power consumption detection model, performing layer-by-layer unsupervised learning from a bottom layer to a top layer in a deep noise reduction self coding network to obtain parameters of encoders and decoders of layers of a network; and setting a Gaussian process classifier at the top layer of the network, performing supervised learning through the data with an abnormal power consumption mark, transmitting an error from the top layer to the bottom layer, and adjusting the parameters of the encoders and decoders of layers of the network. According to the method provided by the invention, power consumption data of multiple platforms are analyzed in combination with the advanced technology in the field of artificial intelligence to mine user power consumption behavior modes hidden in mass data in depth and locate abnormal power consumption suspicious users, therefore abnormal power consumption monitoring is more intelligent and more efficient.