Incorporating Physical Models for Dynamic Stall Prediction Based on Machine Learning