Abstract: Objectives: Aiming at the positioning requirements of vast users in the future, the correlation between the reference stations cannot be removed when the traditional VRS observations are generated, and the interruption of individual reference station data will cause VRS to be unable to be continuously generated. With the increase in the distance between a reference station and a user station, the residual of the spatial correlation error reduces the efficiency of the integer ambiguity resolution and positioning accuracy of the user station. Methods: The undifference VRS algorithm addresses the problem of correlation of the correction values between the station and the satellite. The undifference correction values of each satellite on the reference stations are independent. When the VRS is generated, an optimal reference station subnet can be selected according to the current demand, thus avoiding user ambiguity reconvergence caused by the data interruption of individual reference stations. On the basis of the fixed double-difference integer ambiguity, the reference station of the whole network and the reference satellite's reference integer ambiguity are set to realize the undifference integer ambiguity estimation of the whole network consistency, so as to realize the generation of the undifference error correction value of the whole network consistency. Results: The positioning performance of the undifference VRS algorithm was verified experimentally, four reference stations were selected for the generation of VRS observations, and two small networks were formed in the form of triangulation to fix undifference integer ambiguity, and the spacing of gridded VRS stations was set to 10 km. A virtual reference station R1 was selected to test the undifference VRS algorithm. User stations U1, U2 and U3 were near virtual reference station R1. The generated observations of virtual reference station R1 were used for differential positioning, and the positioning results of users when one of the references could not operate properly were analyzed. 24h of data were used for experiments. The observations of the first 12 hours of R1 were generated by reference station network ABC. When reference station B could not operate normally, the observations of the last 12 hours of R1 were generated by reference station network ACD to test the continuity of observations and user positioning results of the R1. Conclusions: The experimental results indicate that the proposed undifference VRS algorithm can guarantee the continuity of the atmospheric error of the VRS generated during the switching of the reference station network. In adouble-differenceition, the proposed algorithm can also ensure the continuity of the positioning results and achieve centimeter-level high-precision positioning.