Focusing on the problem that most existing time series clustering algorithms based on u-shapelet (unsupervised-shapelet) fail to simultaneously balance the efficiency and quality of u-shapelet extraction, a u-shapelet time series clustering algorithm based on key points guidance named UKey is proposed. Firstly, a subset of time series is selected by random sampling from the time series dataset. A two-step method is proposed to identify the key points in the sampling time series. Then, these key points are utilized to extract subsequences to obtain the u-shapelet candidate set. This strategy not only ensures that the extracted candidate subsequences capture the key fluctuation regions but also effectively reduces the number of candidate subsequences. Subsequently, the Davies-Bouldin (DB) index is introduced as a new quality evaluation method, aiming to ensure that the obtained u-shapelet set exhibits high quality by comprehensively considering inter-class separability and intra-class compactness. Finally, the $ k $-Means is used to cluster the distance matrix constructed based on the u-shapelet set. Experimental results on 10 different datasets demonstrate that the UKey outperforms 14 comparison algorithms, achieving higher accuracy and better interpretability.