Ultra-wide band (UWB) technology has garnered significant attention in the field of indoor positioning due to its high temporal resolution and strong penetration capability. However, traditional UWB positioning methods for non-line-of-sight (NLOS) identification and compensation often fail to accurately characterize channel states in complex environments, leading to insufficient positioning accuracy and precision. This study proposes an autonomous classification approach, termed SimCLR-CIR-SC, which leverages the SimCLR framework for feature extraction from channel impulse response (CIR) data, and combined with the principles of spectral clustering (SC). Based on the autonomous classification results, we designed a time convolutional neural network with attention mechanisms (TCN-A) model to determine channel state categories. For each identified channel state category, a customized TCN-A model is then employed to predict ranging errors. These errors are used to compensate measuring distances and calibrate ranging weights, integrating with the weighted least squares (WLS) algorithm to locate unknown nodes. Experimental results demonstrate that the proposed SimCLR-CIR-SC method effectively and autonomously classifies and labels channel states, outperforming three existing clustering methods. The TCN-A classification model achieved an accuracy of 98.16%, surpassing five existing classification models. Furthermore, the proposed positioning method achieved an average error of 0.57 meters with three anchors, enhancing the positioning accuracy by at least 31.3% compared to four existing methods, and the positioning accuracy improves substantially as the number of anchors increases.