Abstract: Text2SQL is an essential task in natural language processing scientific research. It plays a crucial role in studying intelligent question and answer systems, where the core task is to automatically convert questions described in natural language into SQL query statements. Current research focuses on improving the matching accuracy of SQL clause tasks. However, it ignores the correctness of syntactic generation of SQL, and the production of SQL involving multiple tables joining still suffers from a large number of errors. As a result, a neural network-based Text2SQL approach is proposed, which refactors the database schema to focus on the correctness of SQL syntax generation through an inverse normalization technique called RNSQL (Reverse Normalization SQL). Validated by theoretical analysis and experiments on the public dataset Spider, RNSQL can effectively improve the quality of Text2SQL tasks.