Abstract: Temperature rise of high-speed bearings in gearboxes serves as a critical indicator reflecting the operating condition and health status of 5 MW wind turbines. Aiming at the strong nonlinearity and multi-factor coupling effect affecting high-speed bearing temperature rise, this paper proposes a regression prediction method combining Bayesian-TCN and SHAP analysis for bearing temperature rise. Firstly, a temporal convolutional network regression model is established, and the Bayesian optimization algorithm is adopted to optimize its hyperparameters. Secondly, the SHAP method is introduced to interpret the importance and action mechanism of input features of the model. Finally, simulation experiments are implemented on the MATLAB platform. The results verify that the proposed method achieves favorable prediction accuracy, which can provide references for wind turbine condition monitoring, early fault warning and maintenance decision-making.