Abstract: Sea Surface Temperature (SST) is a key physical variable reflecting ocean-atmosphere interactions, and its accurate prediction is crucial for climate forecasting and disaster early warning. This paper proposes a TimeMixer-based multi-physical-factor fusion prediction model for intelligent SST forecasting at the weather scale in the Banda Sea region. The model first downsamples 30-day historical multivariate time series to capture SST sequences at different temporal scales. Then, a historical data decomposition and fusion module is employed to separate the multi-scale SST sequences into trend and seasonal components, which are modeled using top-down and bottom-up strategies, respectively. Meanwhile, Swin Transformer block is introduced to extract spatial and channel features of the SST sequences. Finally, the model fuses the predictions across multiple temporal scales to generate 7-day SST forecasts. Experimental results in the Banda Sea region show that the proposed model achieves a root mean square error (RMSE) of 0.22 °C, outperforming various baseline models and demonstrating its effectiveness and stability.