With the development of data collection and storage technologies, high-dimensional time series data has be-come increasingly prevalent in various scenarios. Time series data visualization, which intuitively depicts the dynamic changes in such data through graphical representations, has become an important research direction in data science and information visualization. Existing static data visualization methods have limitations, such as inconsistent projection results, while current time series visualization techniques, including Joint t-SNE, suffer from high computational complexity. To address these issues, we propose Joint UMAP. We first introduce a Graphlet Frequency Distribution (GFD) to capture similarities between data points in adjacent time frames and then integrate a new vector constraint into UMAP’s loss function to preserve inherited structures in the high-dimensional data space, ensuring consistency in subsequent projections. Using both synthetic and real-world datasets, we demonstrate that Joint UMAP outperforms existing techniques in terms of Local Coherence Error (LCE) and Cross-Entropy (CE). By eliminating the global dependency of projection results during optimization, Joint UMAP is particularly advantageous in dynamic scenarios where datasets change over time.