As fundamental equipment of advanced metering infrastructure (AMI), smart meters provide system wide visibility for distribution network. However, smart meter measurements often are not synchronized which may cause inaccuracy and divergence when directly applied to distribution system state estimation (DSSE). In order to achieve higher accuracy, the load variation between the near real-time measurements and real-time measurements which is defined as credibility are considered with the historical residential loads taken by high sample rate meters. According to the statistical result, the load variation is exponential distributed at the 95% confidence level which is utilized to update the measurements weight and 95% confidence intervals of state variables are described. A 116-node distribution network is selected to validate the proposed method. And, simulation results demonstrated its accuracy and sensitivity.