Due to complexity and variability of indoor environment and serious non-line of sight（NLOS）and multipath effect

traditional fingerprint positioning technology will cause large positioning errors.Aiming at this problem

using the characteristics of ultra wide band（UWB）signal ranging information accuracy and small fluctuation

using the ranging value as the fingerprint

a three-dimensional indoor localization method based on sparse auto encoder（SAE）and random forest（RF）is proposed.Firstly

more robust eigenvalue are extracted by SAE

and then the eigenvalue are used as the input of the deep neural network（DNN）regression network to obtain the estimated positioning coordinates of the target nodes.At the same time

the old database caused by environmental changes can not match the new acquired fingerprints

and the ranging value is used as the input of the random forest regression model to correct the positioning error of predict positioning coordinates.The experimental results show that the proposed SAE-RF positioning method is more suitable for dynamic and complex indoor environments than the other fingerprint positioning methods and the positioning precision is higher.