The computational accuracy of concrete prediction models is the key to investigate shrinkage influence on performance of concrete structure. Commonly used concrete prediction models are evaluated for their accuracy by comparing their predicted results against the six groups of test data which were collected from published papers. The results show that prediction models were not coincided very well with the experimental data. The phenomenon of concrete structure is a result of several interacting physical mechanisms and is influenced by many variable factors. The shrinkage deformations invariably exhibit large statistical scatter. The calculated results of prediction models could not agree very well with the test results. In order to improve the calculation accuracy of concrete structure prediction models, updating the prediction models based on short-time tests is an effective technology. And the general technology is not proper because of ill-posed problem. So a new improvement technology was proposed and suggested in this study. Seven groups of concrete structure test data were used to evaluate the suggested technology. It could be found that the suggested technology could match better with the test data.