Research On Metal Pipeline Corrosion Early Warning By Integrating Knowledge Graph And Neural Network

Metal pipelines play a crucial role in the transportation of oil and natural gas.However,metal pipelines are prone to corrosion during use,which can lead to a reduction in pipe wall thickness and an increase in the risk of leaks and accidents.In this paper,the metal pipeline in Tahe Oilfield is taken as the research object,and the corrosion data are sorted out by constructing the knowledge map of metal pipeline corrosion field.The artificial neural network is used to predict the pitting corrosion rate of the pipeline,and then the SOM neural network is combined to give the early warning of metal pipeline corrosion.The main achievements of this paper are as follows:（1）The construction of a knowledge graph of corrosion in metal pipelines.The article proposes a BERT-Bi LSTM-CRF joint extraction model as the knowledge extraction model for corrosion data in metal pipelines.Through experimental verification,the model’s accuracy（F1value）in extracting corrosion data from metal pipelines is 91.75%.The extracted knowledge is fused using ontology mapping,and the resulting triple information is stored in a Neo4j graph database,completing the construction of the corrosion knowledge graph in metal pipelines.（2）Development of a corrosion rate prediction model for metal pipelines.The study uses CO₂content,H₂S content,water content,pressure,temperature,flow rate,and p H value as input parameters and point corrosion rate as the output result.BP neural network,RBF neural network,and random forest method are used to predict the point corrosion rate.The BP neural network is found to be the most effective in predicting point corrosion rate when the target error is set at 0.015.The BP neural network model is then optimized using a genetic algorithm to achieve precise prediction of point corrosion rates.（3）Development of a corrosion warning model for metal pipelines.The study employs a SOM neural network as the warning model,using point corrosion rate,nominal wall thickness,service life,temperature,pressure,and flow rate as input parameters.Six sets of common corrosion evaluation data are used as samples,and 1000 training sessions are performed to achieve classification.60 sets of effective sample data are extracted from the knowledge graph for testing.The warning model’s prediction accuracy is found to be 96%after comparing the model’s warning results with the actual results,indicating a high level of accuracy.