Abstract: The underground substation environment is complex, with many electrical equipments, obstacles and walls. The traditional indoor positioning system has low positioning accuracy and cannot meet the needs of daily work. To solve the problem, this paper proposes a non-line-of-sight error suppression algorithm based on the ultra-wideband-Chan -Gauss-Newton iterative -BP neural network (Chan Gauss Newton-BPNN, CGN-BPNN). First, the coordinates obtained by the Chan algorithm are used as the initial values of the Gauss-Newton iterative algorithm to iterate, and the residuals of the algorithm obtained are weighted to increase the weight of the measured data of the base station closer to the label, and then the coordinate value obtained after the iteration is used as the input of the trained BP neural network for correction, and the output is the final positioning result. Computer simulation results show that the positioning accuracy of this algorithm is improved by 42.9% compared with the traditional Chan algorithm, and by 21.9% compared with the Chan Gauss Newton algorithm, which has a wide and positive significance for the application of UWB technology in underground substations.