Several alarm systems with different degrees of autonomy and accuracy have been developed to prevent stuck pipe anomaly, which is common and costs for the oil & gas industry a large amount of the turnover. This thesis proposes a novel data pipeline that starts from drilling rig logs and ends with real-time predictions of the well health status by combining data mining and machine learning techniques. These methods are applied to: overcome several data issues (small amount of stuck pipe examples, lack of important measurements such as inputs of the operator, inconsistencies and so forth), create features and targets for a supervised binary classification problem, train the probabilistic model (which will output the well status), assess its performances. The proposed system, in the tested drilling rigs, outperforms the system currently used and it is in general able to reduce stuck pipe NPT (Non Productive Time) providing net time savings.
Diversi sistemi di allarme con vari gradi di autonomia e accuratezza sono stati sviluppati al fine di prevenire l’anomalia di presa di batteria (stuck pipe), la quale, essendo frequente, costa all’industria petrolifera un’importante porzione del fatturato totale. Questa tesi propone un sistema di elaborazione dei dati che parte dai registri dell’impianto di perforazione e arriva a fornire una misura dello stato di salute del pozzo in tempo reale, combinando tecniche di data mining e machine learning. Tali metodi sono applicati per: risolvere diversi problemi riguardanti i dati (pochi esempi di presa di batteria, mancanza di importanti misurazioni come gli input dell’operatore, inconsistenze e così via), predisporre i dati per un problema di classificazione binaria supervisionata, allenare il modello probabilistico (il quale fornirà predizioni sullo stato del pozzo) ed infine per verificare le sue performance. Relativamente agli impianti di perforazione testati, il sistema proposto supera le performance del sistema attualmente impiegato ed è in generale in grado di ridurre il tempo non produttivo dovuto alla presa di batteria (NPT, non productive time) fornendo un netto risparmio di tempo.
An online system for stuck pipe prediction based on machine learning techniques
PATRUNO, GIOVANNI;VETERE, ALESSANDRO
2016/2017
Abstract
Several alarm systems with different degrees of autonomy and accuracy have been developed to prevent stuck pipe anomaly, which is common and costs for the oil & gas industry a large amount of the turnover. This thesis proposes a novel data pipeline that starts from drilling rig logs and ends with real-time predictions of the well health status by combining data mining and machine learning techniques. These methods are applied to: overcome several data issues (small amount of stuck pipe examples, lack of important measurements such as inputs of the operator, inconsistencies and so forth), create features and targets for a supervised binary classification problem, train the probabilistic model (which will output the well status), assess its performances. The proposed system, in the tested drilling rigs, outperforms the system currently used and it is in general able to reduce stuck pipe NPT (Non Productive Time) providing net time savings.File | Dimensione | Formato | |
---|---|---|---|
2017_12_Patruno_Vetere.pdf
non accessibile
Descrizione: Testo della tesi Revisione 2
Dimensione
7.19 MB
Formato
Adobe PDF
|
7.19 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/138833