Abstract:Based on the GNSS data collected in the urban environment, we evaluate the performance of the unsupervised classification algorithm and the graph optimization algorithm to reduce the multipath errors . The results show that the accuracy of SPP can reach 3.61 m, 2.90 m and 8.14 m in the N, E and U directions based on the K-means++ unsupervised classification algorithm; 53.18%, 55.18% and 44.96% improvements are observed compared to the traditional algorithm. The graph optimization method making use of the pseudorange and doppler factors can achieve an accuracy of 0.94 m, 1.34 m and 2.78 m in the N, E and U directions, with the improvements of 82.1%, 78.5% and 82.0% compared to the traditional method. The graph optimization algorithm is proven to decrease the multipath errors in GNSS positioning under urban environments, which can be used to eliminate the abnormal satellites and provide precise initial coordinates for precise positioning, improving the performance of GNSS positioning in urban environments.