Agharzayeva, Zinyat (2018-08). Application of Machine Learning and Data Analytics Methods to Detect Interference Effects from Offset Wells. Master's Thesis. Thesis uri icon

abstract

  • The goal of this thesis is to demonstrate that linear-based data-driven models are innovative and robust. They have the potential to forecast well bottom-hole pressure and identify interference effects between wells. Permanent Downhole Gauges (PDGs) provide a continuous real-time record of pressure and temperature in the downhole environment. These real-time downhole measurements of pressure contain information about the reservoir properties and interactions with offset wells. This work presents a methodology to reproduce well bottom-hole pressure behavior quickly and to forecast future behavior using those measurements. It also identifies the influence of offset wells based on flowrate-pressure measurements using linear data analysis methods. In this methodology, we chose linear-based machine learning methods as they are much faster, more robust, and more easily interpreted. Furthermore, we formulate the functional relationship between flowrate and bottom-hole pressure into linear relationships using superposition techniques and physical flow behavior assumptions. Then, without making any further physical assumptions, we regulate process into two stages -- training and testing. Training is the regression phase where the flowrates and pressures are correlated using linear machine learning algorithms. Testing is the extrapolation, or forecasting, of the training model to predict well pressure behavior based on a flowrate history. First, to identify offset well interference effects for a selected well, we reproduce the well's bottom-hole pressure response using only flowrate and time data for that well. Subsequently, we test the influence of offset wells on the selected well's bottom-hole pressure response by considering the selected well and offset well's flowrate history one at a time, until we have examined all possible offset wells. By systematically studying the effects of offset wells on the selected well's bottom-hole pressure, we are able to determine the interference of offset wells using only flowrate histories for the considered wells. We validate the methodology by using a synthetic reservoir model whose behavior (connectivity) is known. We reproduce and forecast the pressure behavior of a selected well and determine the influence of offset wells. Then, we compare the identified interference wells with known answers. We note that there is an agreement between the algorithm's results and synthetic model. Also, we test the methodology on the actual field cases. We observe agreement between identified interference effects from offset wells using linear-based data analytics method and those determined from the interpretation of multi-well tests and dynamic observations.
  • The goal of this thesis is to demonstrate that linear-based data-driven models are innovative and robust. They have the potential to forecast well bottom-hole pressure and identify interference effects between wells.

    Permanent Downhole Gauges (PDGs) provide a continuous real-time record of pressure and temperature in the downhole environment. These real-time downhole measurements of pressure contain information about the reservoir properties and interactions with offset wells.

    This work presents a methodology to reproduce well bottom-hole pressure behavior quickly and to forecast future behavior using those measurements. It also identifies the influence of offset wells based on flowrate-pressure measurements using linear data analysis methods.

    In this methodology, we chose linear-based machine learning methods as they are much faster, more robust, and more easily interpreted. Furthermore, we formulate the functional relationship between flowrate and bottom-hole pressure into linear relationships using superposition techniques and physical flow behavior assumptions. Then, without making any further physical assumptions, we regulate process into two stages -- training and testing. Training is the regression phase where the flowrates and pressures are correlated using linear machine learning algorithms. Testing is the extrapolation, or forecasting, of the training model to predict well pressure behavior based on a flowrate history.

    First, to identify offset well interference effects for a selected well, we reproduce the well's bottom-hole pressure response using only flowrate and time data for that well. Subsequently, we test the influence of offset wells on the selected well's bottom-hole pressure response by considering the selected well and offset well's flowrate history one at a time, until we have examined all possible offset wells. By systematically studying the effects of offset wells on the selected well's bottom-hole pressure, we are able to determine the interference of offset wells using only flowrate histories for the considered wells.

    We validate the methodology by using a synthetic reservoir model whose behavior
    (connectivity) is known. We reproduce and forecast the pressure behavior of a selected well and determine the influence of offset wells. Then, we compare the identified interference wells with known answers. We note that there is an agreement between the algorithm's results and synthetic model. Also, we test the methodology on the actual field cases. We observe agreement between identified interference effects from offset wells using linear-based data analytics method and those determined from the interpretation of multi-well tests and dynamic observations.

publication date

  • August 2018