Abstract: Lacking in real-time measurements in distribution system call for adding pseudo measurements and virtual measurements to improve the measurement redundancy, this paper presents a convolutional neural network (CNN) based method for pseudo measurement modeling. Input historical data of different types of load and train the CCN to get pseudo measurements. Then, process virtual measurements in the form of linear constraint through improved equivalent current measurement transformation algorithm for distribution system state estimation (DSSE). The proposed algorithm improves the accuracy and numerical stability of DSSE.