Abstract: Due to the characteristics, such as massiveness, multi-source, diversity and great difference of value density, of the data of natural gas pipeline network, it is of great difficulty to do data analysis and application. With reference to the study results of big data in power grid, supply chain and internet, a big data analysis framework of natural gas pipeline network based on data processing, data mining and comprehensive analysis of multivariate data is put forward. Definitely, the method and function of data processing is illustrated in terms of data cleaning, feature selection and reconstruction. Based on the specific business and scenarios, it is defined that prediction and early warning, model identification, rule learning and deduction are the basis to construct the data mining method of pipeline network. In addition, the comprehensive application of diversified data of pipeline network is discussed. It is also pointed out that the development of multimode learning and federated learning is the key to break the data barrier and to form data intelligence of pipeline network. Through big data analysis of natural gas pipeline network, the big data "ecology" of pipeline network shall be improved continuously, the machine learning method of knowledge in data fusion shall be deeply researched, and a cross-border interpretable and controllable big data analysis method system of pipeline network shall be established, so as to provide theoretical support for development of intelligent pipeline network technologies.