Ultra-wideband (UWB) is considered as a mainstream positioning technology in the field of indoor positioning by virtue of its high-precision positioning, low power consumption and strong penetration. However, indoor positioning still faces a series of challenges, such as NLOS problem. NLOS environments are particularly common in indoor positioning, where signal reflections, refractions, and scattering between buildings and obstacles can lead to performance degradation of traditional positioning algorithms. Aiming at indoor WSN localization scenarios, a SVM-based UWB Indoor Positioning (SUIP) algorithm based on Support Vector Machine (SVM) is proposed, which is applied in a public dataset with different scenarios. In the model optimisation of SUIP, a parameter selection method of SVM is designed through cross-validation. It is experimentally verified that SUIP can reduce positioning error and maintain the robustness efficiently in three-dimension NLOS scenarios compared with classical Chan algorithm.