ABSTRACT

Pore pressure measurements are essential over the life of an oil field, providing critical information from exploration through development. However, designing pressure tests with a formation tester (on either wireline or logging-while-drilling tools) might not be straightforward as it requires understanding several variables such as overbalance, permeability (both horizontal and vertical), viscosity, porosity, and compressibility. Modeling software exists to predict the behavior of a pressure response under multiple combinations of those variables. Typically, trial and error is used to adjust the best pressure test parameters, which are drawdown volume, drawdown rate, and buildup time. The goal is to optimize volume, rate, and time in such a way that a fully stabilized pressure measurement is obtained in the minimum time possible. In practice, this is difficult since each test must be evaluated over the many different variable combinations one expects to encounter. In this work, we develop a method to automate this process to suggest optimal pressure test parameters, given expected distributions of the key variables. This optimization includes a trade-off between the probability a given test would work for a random variable combination and the time each test takes, which is left to the user to decide. Additionally, the user can generate multiple pressure tests, where each is optimized to solve the scenarios the previous test failed to solve.

In this work, we are treating each pressure measurement as two individual tests: an initial drawdown followed by a pressure stabilization time and a subsequent second drawdown and second stabilization time. As each drawdown is defined by a volume and a rate, each full test consists of six parameters. Given expected distributions over the key formation variables, expressed in terms of P10, P50, and P90 values, the goal is to instantly know in what proportion of all scenarios a given test configuration will work. This is done using memoization. A dataset of 1000 scenarios is generated prior to the testing using the given distributions. To each of these scenarios, the pressure responses of different drawdown volumes and rates are simulated and subsequently registered using modeling software. By using the fact that the result of the second test depends on the stabilization pressure of the first, it is possible to disentangle the second test from the first by always stopping the first test at a given pressure within a predefined range. This reduces the combinatorial complexity significantly as the two tests can be analyzed independently. During optimization, it is thus possible to instantly know, without further use of the modeling software, which scenarios will work for a given test configuration, allowing for rapid configuration search and providing the best solution for each total test time. The process can be repeated with unsolved scenarios to build a toolbox of different tests for the different scenarios.

The proposed method is rapid and flexible at testing time and leaves much room for the user to influence the optimization process, focusing on certain parts of the variable space or optimizing with respect to what is unsolved by previous tests. The time-consuming step is the one-time cost of memoization, which for 1000 scenarios takes approximately 48 hours on a standard laptop for 100 volume-rate combos for each drawdown, giving 10,000 possible test configurations to search among. Once the memoization is done, pressure test parameters can be instantly evaluated over all scenarios, and optimization for all test parameter combos can be made in less than a minute. Preliminary analysis confirms that the tests suggested by the method work well in simulation and outperform both the standard tests as well as handcrafted tests by domain experts.

This content is only available via PDF.
You can access this article if you purchase or spend a download.