TY - BOOK

T1 - The application of modern mathematical methods to understand, plan and forecast production cost optimization scenarios in the late field life

AU - Sirghii, Viorica

N1 - no embargo

PY - 2022

Y1 - 2022

N2 - Digitalization has had a significant impact on the complexity of oil and gas operations and has remodeled the entire industry. The ever-increasing amount of generated data accompanied by the energy transition, compliance with contemporary health, safety, and environmental regulations, followed by a cost-efficient production, represent both opportunities and challenges for the oil and gas companies nowadays. Examples of machine learning and artificial intelligence are everywhere around us and have been used for solving complex engineering tasks to bring more efficiency and safety to our daily lives and within major industrial operations. The aim of this research was to develop a novel approach for predicting and preventing failures in wells equipped with artificial lift production systems by using machine learning tools and artificial intelligence algorithms based on the extensive amount of data an oil and gas company is generating every day. Here I created a diagnostic tool for the automatic identification of the sucker rod pump states and malfunctions using digitally generated dynamometer cards. The proposed solution based on artificial neural networks led to a high precision recognition algorithm which can be used in preventing potential well failures and optimized production. Another major task was the identification of trends exhibited by the sucker rod pumps¿ behavior and forecasting future pump states based on the identified trend. Various models have been found, tested and the most fitting approach was selected. The selected model was able to accurately and reliably predict results, almost identical to the real data points. These predictions can be used in daily operations for avoiding potential failures and malfunctions in sucker rod pumps, reduce costs, risks, and increase the mean time between failures. A novel automatic system for detecting and predicting unwanted events in the sucker rod pump operation was created. The results carry high degrees of precision, accuracy, and flexibility, which allow the application and extension of the model to other similar cases. Presented methods and functionalities based on artificial intelligence techniques have demonstrated its power as an enabling technology capable of delivering outstanding outcomes and help solve complex problems.

AB - Digitalization has had a significant impact on the complexity of oil and gas operations and has remodeled the entire industry. The ever-increasing amount of generated data accompanied by the energy transition, compliance with contemporary health, safety, and environmental regulations, followed by a cost-efficient production, represent both opportunities and challenges for the oil and gas companies nowadays. Examples of machine learning and artificial intelligence are everywhere around us and have been used for solving complex engineering tasks to bring more efficiency and safety to our daily lives and within major industrial operations. The aim of this research was to develop a novel approach for predicting and preventing failures in wells equipped with artificial lift production systems by using machine learning tools and artificial intelligence algorithms based on the extensive amount of data an oil and gas company is generating every day. Here I created a diagnostic tool for the automatic identification of the sucker rod pump states and malfunctions using digitally generated dynamometer cards. The proposed solution based on artificial neural networks led to a high precision recognition algorithm which can be used in preventing potential well failures and optimized production. Another major task was the identification of trends exhibited by the sucker rod pumps¿ behavior and forecasting future pump states based on the identified trend. Various models have been found, tested and the most fitting approach was selected. The selected model was able to accurately and reliably predict results, almost identical to the real data points. These predictions can be used in daily operations for avoiding potential failures and malfunctions in sucker rod pumps, reduce costs, risks, and increase the mean time between failures. A novel automatic system for detecting and predicting unwanted events in the sucker rod pump operation was created. The results carry high degrees of precision, accuracy, and flexibility, which allow the application and extension of the model to other similar cases. Presented methods and functionalities based on artificial intelligence techniques have demonstrated its power as an enabling technology capable of delivering outstanding outcomes and help solve complex problems.

KW - machine learning

KW - artificial intelligence

KW - sucker rod pump

KW - dynamometer cards

KW - oil

KW - gas

KW - production

KW - failure prediction

KW - artificial neural networks

KW - trend analysis

KW - digitalization

KW - features

KW - elliptic fourier transform

KW - pump states

KW - Maschinelles Lernen

KW - künstliche Intelligenz

KW - Gestängetiefpumpe

KW - Dynamometerkarten

KW - Öl

KW - Gas

KW - Produktion

KW - Fehlervorhersage

KW - künstliche neuronale Netze

KW - Trendanalyse

KW - Digitalisierung

KW - Features

KW - elliptische Fourier-Transformation

KW - Pumpenzustände

M3 - Doctoral Thesis

ER -