Financial time series forecasting remains a focalpoint of research in finance due to its crucial role in investmentdecision-making and risk management. However, the highlynonlinear and non-stationary characteristics of financial marketspose significant challenges for prediction. This paper introducesa Frequency-Adaptive Normalized (FAN) time series predictionmodel that enhances forecasting accuracy through an innovativefrequency domain analysis approach. The model employs adual-path architecture, incorporating frequency-adaptive nor-malization mechanisms and residual learning strategies, whicheffectively captures both the periodic patterns of time series andaccurately models fine-grained market fluctuations. Experimentsconducted on the TSLA stock dataset demonstrate that theFAN model achieves substantial improvements in both predictiveaccuracy and computational efficiency compared to traditionalmethods. Notably, the model exhibits robust performance whenforecasting during periods of intense volatility. Ablation studiesfurther validate the necessity of each model component, providingnew research directions for financial time series prediction.