Cerebral blood flow velocity (CBFV) reconstruction plays a crucial role in evaluating cerebrovascular function, particularly in the early diagnosis of cerebrovascular diseases, optimizing treatment plans, and preventing strokes. Existing CBFV reconstruction methods face challenges in accuracy and efficiency when processing multivariate time-series signals, particularly in the context of data scarcity and complex signal processing. This study proposes a multivariate time-series model based on a Transformer encoder, which achieves high-precision CBFV reconstruction using arterial blood pressure and CO₂ time-series signals. The model design is based on a long short-term memory module, which effectively compensates for the limitations of the global attention mechanisms in processing local information and enhances local feature learning. Additionally, a hybrid loss function is employed to optimize local waveform errors, improving reconstruction accuracy. Furthermore, to address the issue of data scarcity in the target domain, this study introduces a transfer learning strategy based on the correlation between arterial blood pressure and electrocardiogram signals, alleviating the impact of limited data on model performance. Experimental results demonstrate that the proposed model outperforms traditional regression and deep learning models in the CBFV reconstruction task, with a Pearson correlation coefficient of 0.51870, a dynamic time warping distance of 17.879, and mutual information of 0.34375, while completing the reconstruction of 200 data points in 0.04 s. The study validates the effectiveness of this method in precision medicine and provides innovative solutions for clinical diagnosis, disease prevention, and personalized treatment, with broad application prospects, particularly in medical signal processing, intelligent healthcare, and health monitoring.