Enhanced Detection Method for Minor Leaks in Process Pipelines Based on Ensemble Learning

Abstract

Seeking to address the shortcomings of conventional pipeline leak detection techniques marked by high hardware costs, challenges in real-time monitoring, difficulties in constructing effective detection models, and notable inaccuracies, a novel cost-efficient, high-precision, real-time weak leak detection method (ensemble learning weak leak early warning model) for gas pipelines is proposed based on ensemble learning. Initially, the pressure and flow data of the pipeline inlet and outlet are captured by strategically placed pressure and flow sensors. Subsequently, the long short-term memory (LSTM), multivariate state estimation technique (MSET), and multiple regression modeling of real-time transients (MR-RTTM) models are selected as foundational models, with specific leakage alarm thresholds established based on the distinct characteristics of each model. These thresholds facilitate the preliminary assessment of pipeline operational status. Furthermore, a multitier pipeline weak leak alarm strategy is devised, and ensemble learning is employed to culminate in an ensemble learning-based weak leak detection methodology. The model is validated through experiments conducted on a pipeline leak test bench. The results demonstrate that compared with individual methods such as LSTM, MR-RTTM, and MSET, the ensemble learning model maintains an overall leakage detection rate above 94%, sustains a false alarm rate below 0.4%, and achieves a mean accuracy of 96.92%. Under noisy conditions, the maximum detection rate for 0.5% minor leakage remains at 99.94% with the ensemble model, while the minimum false alarm rate reaches 0.042%, maintaining evaluation accuracy above 85% and an average alarm delay within 8.4 s for 0.5% leakage rate scenarios. The novel ensemble-learning-based pipeline leak detection method proposed in this paper offers a new approach to enhancing the detection capability of weak leaks in process pipelines within oil and gas stations.