Deep learning is a new field of machine learning. Its powerful data analysis

prediction

and classification capabilities satisfy the needs of big data applications in smart grid. Firstly

this paper summarizes the basic ideas of deep learning

introduces the structures

basic principles and training methods of five models of deep learning(generative adversarial network

recurrent neural network

convolution neural network

stacked auto encoder and deep belief network)

and summarizes their application characteristics. The applications of deep learning techniques in fault diagnosis

transient stability analysis

load and new energy power forecasting

operation control in power system are summarized. Based on the technical characteristics of deep learning and the production links of power system

the application framework of deep learning technology in power system is constructed. Finally

the application of deep learning is prospected in the aspects of multi-energy system operation regulation

power electronic system security analysis

flexible equipment fault diagnosis

and cyber-physical power system security protection.