Application of artificial intelligence techniques to well spacing optimization in fractured tight reservoirs

Well spacing optimization is critical to the success of any hydrocarbon field development particularly unconventional fields and significantly affects the economic performance of oil and gas production operations. Oil and gas field operators over the years have resorted to field trials, surveillance and numerical simulation as the modes of determining optimal well spacing. Although field trials and surveillance are relatively easier to undertake, they are characterized by huge capital investments and a large amount of time requirement in order to arrive at conclusive decisions. Additionally, there are huge number of combinations to test, making it difficult to find a definitive solution. The use of numerical simulation is also computationally expensive in exploring various combinations of field development schemes (well spacing and completion design). As such, in spite of the many theoretical and practical studies, the desire to achieve a fast-paced well spacing optimization outcome is not yet realised. In this study, artificial intelligence techniques are employed with the primary aim of developing a workflow for optimizing well spacing in unconventional fractured tight reservoirs. To achieve this objective, a coupled artificial neural network (ANN) based proxy and evolutionary algorithm is proposed and applied to Avon Hill field in west central Saskatchewan which is a tight fractured oil field. Multilayer perceptron (MLP) neural network, particle swarm optimization (PSO) and genetic algorithm (GA) were utilized. In the development of the ANN based proxy model, four input parameters (well spacing, fracture half-length, fracture conductivity and production time) and two outputs (cumulative oil and gas production) were used. The training and testing data were obtained from a history matched numerical reservoir model of the Avon Hill field. The ANN topology consists of one hidden layer with 20 neurons. Logarithmic sigmoid (logsig) activation function and Levenberg–Marquardt backpropagation training algorithm were employed. The trained proxy model is then coupled with PSO and GA and serves as a substitute for the numerical simulation in the optimization process. Maximizing net present value (NPV) was considered as the objective function while well spacing, fracture half-length and fracture conductivity were set as the optimization variables. 3 well spacing optimization workflows are then developed comprising of ANN based proxy only, ANN-PSO and ANN-GA. It is found that the use of the ANN based proxy model enables various combinations of well spacing and fracture designs to be tested at higher computational speed compared to the numerical simulation. Comparing the two evolutionary algorithms used, the results show that the PSO outperformed the GA in terms of converging at a higher objective function value (NPV) and at a higher convergence speed. In relation to the field understudy, based on a 20-year production forecast, the workflow using only ANN based proxy gave NPV of $ 1.38 x 107 which corresponds to 11.3% increase in NPV compared to the base case of $ 1.24 x107. NPV of $ 1.542 x107 was achieved using ANN-PSO which is 24.4% above the base case. The ANN-GA gave an NPV of $1.509 x107 which corresponds to 21.7% increase. The optimal well spacing ranges between 124 m to 132 m. From the results obtained, to achieve a maximized NPV for the field, it is required to drill 6 wells (compared to current 5 wells) and this result is consistently predicted by all 3 workflows used in this research.

Name thesis:degree_name

Master of Applied Science (MASc)

Level thesis:degree_level

Master's

Discipline thesis:degree_discipline

Engineering - Petroleum Systems

Grantor dc:publisher

Faculty of Graduate Studies and Research, University of Regina

Year dc:date.issued

2022

Author dc:creator

• Anumah, Prosper

Advisor dc:contributor.advisor

• Azadbakht, Saman

Committee member dc:contributor.committeemember

• Zeng, Fanhua

Language dc:language.iso

en

DOI

https://doi.org/10.82465/4608

OAI identifier oai:identifier

oai:uregina.scholaris.ca:10294/16044