We create two models to predict the results. The first one is a statistical model and it is more like a draft to test if our hypotheses of time series characteristics are correct or not. Besides, the second one is a machine learning model based on lightGBM which achieves the result measured by RMSE=0.09763
Industrial AI seems to be much more difficult than other fields because of the complicated mechanism of each particular industry scenario and the highly noised/missing data. Learning methods like few-shot transfer learning and zero-shot learning might have its place in these areas.