Research on Leak Detection Algorithms Based on CNN-LSTM Neural Networks

Abstract

In the application of long-distance transmission pipeline, the explicit mathematical model method may not completely obtain all accurate values since the parameters along the line cannot be measured point by point. A practical method based on CNN-LSTM neural network was proposed. This method uses CNN network to find spatial features and uses LSTM to explore temporal features. The real data collected from operating pipeline are used to train and verify the deep learning model, and then use the model to predict the flow rate, with an error range from 0.3% to 0.7% of main line's flow rate. By continuously comparing the actual with the predicted values in real time, a pipeline leak can be found. In addition, an improved location algorithm based on the curve distance between relevant pressure points was proposed. The field test on actual pipeline shows that the proposed new algorithm has reliable performance and does not generate false alarms during pressure fluctuation such as the operation of pipeline equipment.

Keywords

• pipeline leak detection
• cnn
• lstm
• pressure gradient