With the high penetration of distributed generators, the radial structure of conventional distribution network system will change to a multi-terminal type, the traditional fault location method will be invalid. In this paper, a fault location method based on deep learning for active distribution network is proposed. This method firstly collects the current and voltage data through the feeder terminal unit. Combining the power output data, a fault data vector is formed; secondly, it uses tensorflow framework to build a deep neural network model based on fully connected network to mine the mapping relations between fault data vectors and fault sections and form the final fault location model through training. Finally, the fault location results demonstrate the effectiveness of the proposed method. Case studies show that compared with the traditional BP and learning vector quantification neural network model, the deep learning model has faster convergence speed and higher fault location accuracy. The final model has high fault tolerance to information distortion and loss.