Artificial Neural Network (ANN) to estimate reservoir parameters in Naturally Fractured Reservoirs using well test data

Abstract

Abstract Artificial Neural Network (ANN) is used to estimate reservoir parameters from well test data. Theoretical pressure derivative curves, generated by the Sthefest numerical Laplace inverse algorithm and belonged to a Naturally Fractured Reservoir (NFR) with Pseudo Steady State (PSS) inetrporosity flow, used to train the ANN. Instead of using normalized pressure derivative data as ANN׳s input, the coefficients of interpolating Chebysehv polynomials on the pressure derivative data in log–log plot were used as input to the ANN. Different training algorithms used to train the ANN and optimum number of neurons for each algorithm were obtained through minimizing Mean Relative Error (MRE) over test data. According to the results of MRE, it is concluded that Levenberg–Marquardt algorithm has the lowest possible MRE among various training algorithms used to train the ANN. In order to compare the accuracy of the new proposed method with the previously used methods, normalized pressure derivative data and coefficients of the conventional polynomials are used to train the ANN. Results show that employing the coefficients of the Chebyshev polynomials as input to ANN׳s decreases MRE over test data while using coefficients of conventional polynomials worsens learning phase of the neural network compared to the normalization method. The lower values of the MRE when using conventional polynomials arise from the fact that interpolating a function using Chebyshev polynomials is more accurate than other interpolation techniques for a specific number of basic functions. Test data are used to verify the accuracy of the trained ANN to estimate each reservoir parameter. Low relative error values of reservoir parameters when using test data shows the capability of the new method which employs the Chebyshev polynomial coefficients as input to train ANN. Moreover, a field data for a dual-porosity fractured reservoir is applied to make a comparison of ANN׳s outputs with the results obtained using different well test softwares and normalization method. The results show a good consistency between ANN and well test softwares.