Study On The Construction And Application Of Coalbed Methane Knowledge Graph Based On Multi-source Heterogeneous Data

Coalbed methane geology is a typical data-intensive science,with the distinctive "4V" characteristics of large data volume,various types,high velocity,and low-value density.These characteristics lead to the problems of data confusion and missing mechanism in the integration,sharing,mining and analysis of related data.The Knowledge Graph,with its powerful semantic processing and open organization capabilities,provides an effective tool for intelligent organization and application of massive data in the era of Big Data.Coalbed methane knowledge graphs and graph neural networks can be applied with their data and architecture characteristics to promote artificial intelligence research in coalbed methane science in a data-and knowledge-driven research mode.Therefore,conducting research on constructing and applying coalbed methane knowledge graphs based on multi-source heterogeneous data is significant.The main contents and results of this study are as follows:(1)This study developed a top-down and bottom-up knowledge graph construction process based on the knowledge characteristics of CBM disciplines.The ontology layer was developed based on expert opinions,national classification standards,and actual data.Two sets of deep learning models are then used to extract information from the domain’s most widely stored text and image data.In text recognition,Dic BERT-Bi LSTM-CRF is proposed based on professional dictionaries with TF-IDF and BERT-Bi LSTM-CRF.This model can solve the problem of the insensitivity of traditional models to the professional lexicon and improve recognition accuracy.In image information extraction,the combination of DBNet+CRNN is used to complete the text detection and recognition in the image.Then knowledge fusion is performed according to the vector similarity of entities,and finally,the knowledge graph is stored and displayed by Neo4 j.(2)This study proposed a CBM production prediction model with temporal,spatial,and geological features based on knowledge graph and graphical neural networks,T-DGCN,which innovatively uses DTW(Dynamic Time Warping)to measure inter-well similarity and aggregates with geological and spatial features to achieve dynamic correction and dense connection of matrices in a multilayer neural network.The model then uses GRU(Gate Recurrent Unit)to extract temporal features from the gas production sequence and predict daily gas production.Tests on the relevant dataset show that T-DGCN achieves an accuracy of 0.9298 in short-term prediction,which is significantly higher than that of the baseline model.(3)This thesis built a CBM intelligent Q&A system based on Flask,Java Script,Neo4 j,and the UNIT dialogue platform.The system integrates the extended trained Dic BERT-Bi LSTM-CRF model and provides the intention recognition function.The system can realize keyword search,intelligent Q&A,query result display,yield prediction(T-DGCN),and other functions,providing research case for intelligent interaction and deep application of knowledge graph.