Abstract: To deal with the key problems of non Gaussian noise, anomaly and missing in the measured data in distribution network state estimation (DNSE). This paper presents a robust forecast-aided state estimation method (FASE) on distribution network based on the long- short- term memory and particle filter (LSTM-PF) to realize the real-time dynamic state estimation of distribution network. A deep-seated LSTM is employed to establish the prediction model based on the historical operation data of distribution network, and the nonlinear model between measurement and state is constructed by the improved PF to reformulate nonlinear equations. An isolation forest anomaly detection technology is proposed to accurately identify the bad data in the measurement information. The pseudo measurements obtained by deep-seated LSTM model combining with the power flow calculation can correct the missing and abnormal data. In addition, the LSTM-PF algorithm proposed in this paper can predict and estimate the state of nodes after topology changes. Case studies on the IEEE 33-bus system and 10 kV 78-bus real distribution network system verify the high accuracy and robustness of the proposed LSTM-PF method. This paper can provide a reference for the state estimation of DNSE.