Aiming at the characteristics of high complexity and high accuracy of underground coal mine positioning. For improving the accuracy and efficiency of coal mine precise positioning and ensure the safety of personnel, an ultra wideband positioning algorithm based on improved TDOA is proposed. According to the different conditions of underground roadway, different training and filtering methods are adopted to ensure the accuracy and low complexity of underground positioning. The experimental results show that the average positioning accuracy of the combined algorithm can be maintained below 10cm in both LOS and NLOS, and compared with the basic TDOA algorithm, the average positioning accuracy is improved by 20cm and 35cm in LOS and NLOS, respectively; while in Los, the trained neural network is used for high-speed positioning to improve the positioning efficiency. Therefore, the combined algorithm can achieve the balance of positioning accuracy and efficiency in different environments, meet the requirements of high-precision positioning underground, and is suitable for underground personnel positioning.