Abstract: To monitor the force state of the main shaft of wind turbine in real time, the paper constructs an adaptive update prediction model that integrates mechanism and data. Firstly, adopting a "physical-baseline plus deviation-compensation" strategy, the finite element mechanism model is used to provide the prediction reference value , and the data-driven model based on the measured data is combined to compensate for the deviation, thereby overcoming the limitations of a single model. Secondly, an incremental learning method with a variable forgetting factor is implemented. This mechanism enables online learning of new data to update the network parameters, thereby alleviating the performance degradation problem during long-term operation. Validation results demonstrate that the prediction error of the main shaft internal force is stable within 3%, and the update mechanism ensures the long-term prediction accuracy.