<p>Data streams evolve continuously with their distributions potentially shifting unexpectedly, known as concept drift. The scarcity of new distribution samples after concept drift challenges model's distribution capture and rapid adaptation, resulting in poor performance. In this work, we propose the Dynamic Memory Window and Data Augmentation-based Incremental Broad Learning System (DMWDA-IBLS). A dynamic memory window strategy is employed to detect concept drift and isolate storage of historical data. Upon drift detection, historical data distributionally similar to new data are retrieved through a window merging detection. Concurrently, to alleviate the scarcity of new samples, we generate augmented samples faithful to current distribution using three adaptive data augmentation strategies, which dynamically adjusts parameters based on window statistical features. By integrating these relevant historical and augmented samples with new data, the model can capture more valuable information, thereby improving its ability to update effectively. Results demonstrate that DMWDA-IBLS achieves rapid responsiveness to concept drift across real-world and synthetic data sequences, effectively mitigating the difficulty of adapting quickly to concept drift when new samples are scarce.</p>