In view of the similarities and differences in data composition, data accuracy and refresh frequency of different measurement data in distribution network state estimation, a new estimation fusion system is proposed on the premise of ensuring the structure of the traditional state estimator. At the same time, the improved generalized maximum likelihood (GM) estimation and estimation fusion system are combined to estimate the system node voltage amplitude and phase angle. Firstly, GM estimation is used to enhance the robustness of the estimation model. By using adaptive project statistics and analyzing the weight function of the objective function in GM estimation, the improved GM estimation method is applied to the state estimation. Secondly, considering that the traditional measurement system is different from the phasor measurement system in terms of measurement channel and instrument sampling rate, based on the traditional state estimator, the phasor measurement data is fully utilized to estimate the state of different estimation modules. At the same time, the multi-sensor data fusion theory (MDF) is used to fuse the estimated results to obtain the optimal estimation value. Finally, the simulative analysis on the improved IEEE 14 and IEEE 33-bus distribution network examples verifies the validity and reliability of the improved GM estimation and estimation fusion system.