Abstract: In view of the frequent changes of distribution network topology and the difficulty of obtaining the topology structure in real time, a distribution network topology identification method based on attention mechanism and convolutional neural network(ACNN) is proposed. The convolution neural network(CNN) is used to mine the relationship between measurement information and distribution network topology, and learn its mapping rules; Considering the problem of insufficient number of advanced measurement devices such as phasor measurement unit(PMU) and mico phasor measurement unit(μPMU) installed in the current distribution network, the attention mechanism is integrated into the hidden layer of convolutional neural network to enhance the robustness of the model; The dimension of feature data set is reduced by random forest algorithm to reduce the time and space complexity of the model; Finally, numerical examples are carried out based on ieee33 node distribution network and PG & E69 node distribution network to verify the feasibility and superiority of the method, and to test the possibility of topology identification using fewer features. The results show that the proposed method has good superiority, robustness and strong generalization ability, Distribution network topology identification can be realized when only a small amount of time section measurement data is provided, and it is also applicable to radial network and ring network.