As deepwater drilling advances into increasingly narrow pressure margins, the available response time to well-control anomalies diminishes to just a few minutes. Conventional surface-only monitoring techniques often experience signal delays and attenuation along the drillstring, which hampers the early detection of disturbances near the drill bit. This study introduces a dual-point near-bit diagnostic framework designed to provide early anomaly warnings. Specifically, two downhole measurement locations are utilized to capture both synchronous responses and the transmission characteristics between points. A data-driven multi-indicator analytic hierarchy process (MI-AHP) is applied to derive interpretable feature weights by integrating multiple importance metrics obtained from an offline random forest model, facilitating efficient linear scoring. To address class imbalance, a two-stage cascade model is implemented: the first stage discriminates anomalies from seven normal conditions, while the second stage classifies four distinct anomaly types (kick, lost circulation, washout, and bit sticking). The proposed system was validated on data from fifteen deepwater wells, encompassing 1586.3 h and 22 anomaly events, achieving a Macro-F1 score of 93.4% for normal condition recognition, 91.2% for anomaly detection, and 89.8% for anomaly classification. Deployment on an ARM Cortex-M4 embedded platform requires only 78 ms and 125 KB of memory, supporting real-time downhole operation. In independent well testing, the system attained a Macro-F1 score of 90.5% with a false positive rate below 0.6%, successfully detecting a kick event 36 min prior to conventional surface-based identification methods.
扫码关注我们
求助内容:
应助结果提醒方式:
