With increasing attention paid to house safety, automatic health monitoring of old buildings becomes a necessity. In this context, the preprocessor of automatic monitoring data was optimized and a preprocessor was built leveraging Python's abundant "library" resources. The data was screened according to the credibility through Grey Relational Analysis. The abnormal data was screened through Box-plot. Then, the abnormal data and missing data were interpolated and replaced. Finally, the data was smoothed by Kalman filtering so that the dispersion was reduced. Given automatic monitoring equipment was sensitive to surrounding environment, a data comparison method to check automatic monitoring data with manual monitoring data was proposed. This method was aimed at ensuring the accuracy and reliability of automatic monitoring data. The short-term deformation monitoring data was predicted with fitted curve. The prediction accuracy reaches 0.1% and the effective prediction time is about 10 days. This result can meet the need of researching.