Time series contains subsequence patterns of unknown lengths and expressions. Extracting these patterns accurately is essential for uncovering the information within the time series. To this end, we propose a clustering framework to determine the subsequence lengths and segmentation positions adaptively by minimizing intra-class error under the time series cover constraints. Besides, we introduce a virtual noise cluster in our framework to differentiate noise data from normal subsequence patterns, reducing the impact of noise data on the extraction of time series subsequence patterns. We also employ a nested optimization algorithm with a dynamic optimization routine and three optimization operations to refine the subsequence patterns: combining patterns that tend to appear adjacent to each other, splitting patterns with large intra-class errors, and removing clusters when the algorithm gets stuck in local optima. We compare our method with state-of-the-art methods on synthetic and real datasets. Both quantitative and qualitative results demonstrate the effectiveness of our model.