The invention discloses an intelligent access system based on natural language understanding and a method thereof. The system comprises a user natural language interaction module, a Search Enginine, aSchema Graph, a Tokenizer, a Tree Builder, and a SQL Generation. The method comprise: the user natural language interaction module transmitting a natural language to the external word device for external segmentation to separate words, the Tokenizer identifying the meaning of each separated word and combining the words into phrases, and finally, automatically or interactively selecting the optimal combination and explanation with a user; constructing a Query Tree by the Tree Builder according to the node sequence in the Tokenizer, the SQL Generation converting the Query Tree into SQL, sendingthe SQL to a database DB, and feeding back the SQL to a natural language interaction module of the user for reference of the user. The key points of the technical scheme are that more question formsand richer complex queries can be supported, and higher accuracy can be achieved on an MAS data set under the condition of no interaction with a user by using less information.