The high noise and nonlinear characteristics of financial time series data are becoming more and more obvious, and the accuracy of traditional prediction models in stock price forecasting is facing challenges. This paper proposes a hybrid framework based on Variational Mode Decomposition (VMD) and Transformer models to improve the predictive performance of stock opening and closing prices. Firstly, the original index sequence is decomposed by the VMD algorithm, and multi-scale Intrinsic Mode Functions (IMF) is extracted to effectively separate noise and trend information. Secondly, the global self-attention mechanism and parallel computing advantages of Transformer model are utilized to conduct time series modeling for de-noised components and capture long—and short-term dependencies. Finally, the case study selects the daily opening and closing price of Shanghai Composite Index from 2012 to 2024, and the results show that the model in this paper has excellent performance in the root- Mean Absolute Error (MAE), Mean Square Error (RMSE) mean absolute percentage error (MAPE) and Coefficient of Determination (R2) of prediction at different time scales. This study provides a hybrid method with strong interpretation for the high-precision prediction of complex financial data. It has practical reference value for quantitative investment and risk management.