Abstract:To address the challenges faced in the classification of high-dimensional time series data, such as significant noise, dimensional redundancy, and difficulty in capturing long-range dependencies, this paper proposes a novel classification method based on the Transformer architecture. The study first reduces data noise and unifies feature scales through moving average filtering and normalization. It then employs techniques such as Principal Component Analysis (PCA) for data segmentation and dimensionality reduction to enhance computational efficiency. Building on this foundation, the self-attention mechanism of the Transformer model is utilized to effectively capture long-term dependencies within time series, thereby strengthening the model''s ability to understand complex patterns. Experimental results show that the proposed method achieves excellent classification performance on benchmark datasets. Compared to traditional methods, the proposed approach not only improves classification accuracy but also significantly accelerates model training speed. This research provides an efficient and reliable solution for the effective classification of high-dimensional time series data.