(430b) Data-Driven Optimization of Highly Constrained Oil Recovery Processes Using Neural Network Surrogate Models and Classification Based Implicit Constraint Handling Schemes

Crude oil production is a critical component in the global energy sector. As the discovery of new oil fields is becoming increasingly challenging, extending existing reserves through oil recovery is becoming equally important [1]. Oil companies employ waterflooding as a secondary oil recovery technique to release trapped oil and improve the recoverable oil amount by injecting water into reservoirs [2-3]. Globally, 250 million barrels of water are produced per day to recover 80 million barrels of oil per day from depleting reservoirs (an average of three barrels of water for each barrel of oil produced) [4]. As a result, more than $40 billion is spent every year to deal with unwanted water [4]. Optimizing waterflooding can decrease water use, lower operating costs, and increase oil field life, which in turn assists in the sustainability and profitability of crude oil production. However, due to the complex nature of the reservoir characteristics and thousands of process constraints involved, deterministic optimization of secondary oil recovery becomes computationally intractable.

In this work, we present a data-driven optimization framework where highly accurate feedforward neural network (FFNN) surrogate models and classification-based feasibility analysis are employed to address the computational limitations of highly expensive simulations while providing feasible superior solutions to the original problem. We formulate the problem using a well-pressure control approach to maximize the net present value (NPV) of reservoir operations. First, we collect samples from the reservoir simulation offline and train a highly accurate deep FFNN (blind testing R² > 0.90) with sigmoid activation functions and 4 hidden layers using 10,000 well bottom hole pressure samples and their corresponding NPV outputs. The validated and tested neural network then serves as the reservoir simulator to carry out data-driven optimization. For constraint handling, we incorporate a classification algorithm to model and filter out the infeasible samples. This approach was previously shown to be an effective strategy when several constraints are present in the formulation [5,6]. Here, we test this implicit classification modeling approach on thousands of reservoir constraints and make comparisons to the case where these were modeled explicitly with individual surrogate models [7]. We train C-parametrized Support Vector Machines with radial basis kernel using the constraint violation data of the collected samples and classify inputs according to their predicted feasibility. We demonstrate our approach on two benchmark reservoir simulation studies, the Egg model [8] and the UNISIM model [9], and test the performance of local and global data-driven optimization algorithms in maximizing the NPV of reservoir operations. Our results show that significant computational savings can be achieved by integrating regression and classification-based surrogate models in data-driven optimization of computationally expensive simulations where superior objective function values are obtained compared to explicit constraint handling strategies.

References

1. Muggeridge, A., Cockin, A., Webb, K., Frampton, H., Collins, I., Moulds, T., and Salino, P., Recovery rates, enhanced oil recovery and technological limits. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 2014, Volume 372, 20120320.

2. Ahmed, T., Principles of Waterflooding. Reservoir engineering handbook, 2018, 901-1093.

3. United States Department of Energy, Office of Fossil Energy: Enhanced Oil Recovery, (n.d.). http://energy.gov/fe/science­innovation/oil­gas­research/enhanced­oil­r…

4. Bailey, B., Crabtree, M., Tyrie, J., Elphick, J., Kuchuk, F., Romano, C., and Roodhart, L., Water control. Oilfield review, 2000, Volume 12, 30-51.

5. Beykal, B., Aghayev, Z., Onel, O., Onel, M. and Pistikopoulos, E.N., 2022. Data-driven Stochastic Optimization of Numerically Infeasible Differential Algebraic Equations: An Application to the Steam Cracking Process. In Computer Aided Chemical Engineering(Vol. 49, pp. 1579-1584). Elsevier.

6. Dias, L.S. and Ierapetritou, M.G., 2019. Data-driven feasibility analysis for the integration of planning and scheduling problems. Optimization and Engineering, 20, pp.1029-1066.

7. Beykal, B., Boukouvala, F., Floudas, C.A., Sorek, N., Zalavadia, H. and Gildin, E., 2018. Global optimization of grey-box computational systems using surrogate functions and application to highly constrained oil-field operations. Computers & Chemical Engineering, 114, pp.99-110.

8. Jansen, J.D., Fonseca, R.M., Kahrobaei, S., Siraj, M.M., Van Essen, G.M. and Van den Hof, P.M.J., 2014. The egg model–a geological ensemble for reservoir simulation. Geoscience Data Journal, 1(2), pp.192-195.

9. Gaspar, A.T., Avansi, G.D., Maschio, C., Santos, A.A. and Schiozer, D.J., 2016, June. UNISIM-IM: Benchmark case proposal for oil reservoir management decision-making. In SPE Trinidad and Tobago Section Energy Resources Conference. OnePetro.