The invention discloses an electric load power consumption decomposition method based on deep learning. The electric load power consumption decomposition method comprises five steps of: step 1, acquiring a complete working curve about one or several kinds of electrical parameters of all target electrical equipment in a target electricity utilization scenario according to needs; step 2, synthesizing simulation data which simulates the mixed use condition of the electrical equipment in the target electricity utilization scenario according to actual electricity utilization condition in the target scenario by utilizing the complete working curve of the electrical equipment acquired in the step 1; step 3, establishing deep neural networks used for electric load decomposition, and determining a number of neural network layers and neurons contained in each layer as well as a link relationship among the neurons contained in the layers; step 4, establishing a training data set by utilizing the simulation data synthesized in the step 2, so as to train the deep neural networks established in the step 3; step 5, and carrying out online application of the deep neural networks obtained through training. The electric load power consumption decomposition method based on deep learning belongs to the field of electric load power consumption detail monitoring, and can realize the non-invasive estimation of the power consumption of the target electrical equipment.