Abstract: Traditional code review for substation engineering projects relies on manual review based on 2D drawings, which has many drawbacks such as low efficiency and high error rates. By leveraging emerging BIM technology and the IFC international standard for data sharing and interaction in the construction field, this paper takes the review of civil engineering sections as an example to study the implementation technology of machine learning-based automatic code checking for BIM projects in substation engineering. Firstly, the analysis of IFC file is realized by programming. For its decentralized hierarchical structure, locate instances at all levels from top to bottom and integrate related information. The civil engineering information of substation project is extracted, and the extracted data information is stored in SQL Server database in a specific format. Secondly, considering that the specification provisions are numerous and the logic is complex, the LSTM neural network is used to automatically generate the rule structure and encode the relevant civil engineering specification provisions into a rule file. Finally, the Drools.NET rule engine is used to implement automatic civil engineering code checking by utilizing the parsed civil engineering information of substation projects from IFC files and the encoded rule files.