An Integrated Model-Driven and Data-Driven Method Based on CFD-PSO-RBF: Applications to Leakage Detection of Natural Gas Pipelines

To address the issues of low accuracy, poor interpretability, and over-reliance on data in conventional model-based and data-based methods for pipeline leakage detection (PLD), an integrated model-driven and data-driven approach is proposed in this paper. In the presented method, the particle swarm optimised radial basis function (PSO-RBF) neural network is introduced into the computational fluid dynamics (CFD) model of the pipeline. Such the hybrid model improves the accuracy of the pipeline description by employing PSO-RBF to compensate for the errors generated by the model-based approach due to inaccurate meshing and imprecise modeling. The CFD model for the natural gas pipeline is initially built using the Ansys Fluent platform for offline training. It is then integrated with the PSO-RBF model. In the online detection stage, the trained CFD-PSO-RBF model is utilized for online prediction and real-time monitoring of pipeline operation. To demonstrate the application potential of the developed method, it is successfully employed to detect natural gas leakage in a laboratory simulating a real natural gas pipeline. Experimental results show that, in comparison to the traditional methods, the CFD-PSO-RBF method improves the accuracy of leakage detection and performs well in terms of response time.