Search for a command to run...
Abstract This study introduces a novel leak detection methodology for liquid pipelines, combining a Sequential Probability Ratio Test (SPRT) with a Kantorovich Distance (KD) filter (SPRT-KD). This model enhances leak detection performance over traditional volume balance (VB) systems in the traditional domains of reliability, sensitivity, robustness, and accuracy. The model performs well in complex pipeline networks in real time, addressing limitations of conventional methods under transient conditions. The SPRT-KD integrates a statistical data-driven SPRT model based on volume imbalance for statistical leak detection with a KD filter to detect operational changes. SPRT accumulates evidence of leaks via flow discrepancies, employing slow, medium, and fast time window detection filters, while KD measures shifts in the flow distribution to detect and suppress the effect of transients which could cause false alarms. The process involves continuous monitoring of flow, KD-based anomaly detection, and SPRT score aggregation over short period (15-minute) windows to trigger alarms when thresholds are exceeded. In a case study from a Northern Alberta condensate pipeline (December 2023), the SPRT-KD model detected a 40 m3/day leak (0.6% of 6700 m3/day total flow) in 13 minutes, compared to 3 hours for a VB model, despite high transience in the complicated network. The KD filter eliminated false alarm signals attributed to operational changes. The model maintained signal integrity, with no loss of alarm signal over the entire 10 hour leak period. SPRT-KD demonstrated superior sensitivity (detecting leaks below the flow meter accuracy) and speed compared to a VB model. The SPRT-KD model combines a sensitive data-driven model with the use of a real time rolling KD calculation to identify operational changes. This approach minimizes false alarms in real-world environments, even in highly transient networks with a plurality of inlets and outlets. SPRT-KD offers a significant advancement in statistical leak detection, providing rapid, reliable alerts in transient environments, validated by real-world application.