Optimized design of cyclic pressure pulsing in naturally fractured reservoirs using neural-network based proxy models

Stimulation of oil wells with cyclic pressure pulsing using gases is an IOR method that is effectively applicable to fractured reservoirs. Fractures provide a large contact area for the injected gas to diffuse through and penetrate into the low-permeability matrix. Also, high permeability of the fracture system provides easy delivery of the injected gas and produced oil. The process attracts operators because of shorter payback periods and lower investment requirements as compared to field-scale flooding projects. Designing a cyclic pressure pulsing project requires optimization of the design parameters such as the amount of injected gas, injection rate, and lengths of injection, soaking, and production periods. Due to the computational cost of simulating a large number of scenarios, it is an arduous task to determine the aforementioned design parameters, and to optimize the process. In this study, neural-network based proxy models are used to assess the feasibility of these processes in reservoirs with different characteristics, and to develop optimized design schemes to maximize the efficiency of the process. The method of approach includes understanding the mechanics of the process via reservoir simulation studies. Artificial neural network (ANN) based proxy models that accurately mimic the reservoir model efficiently in terms of the computational time are developed. First, the methodology is tested with the reservoir model of the Big Andy Field where cyclic CO2 (since 1985) and cyclic N 2 (since 1996) injection have been utilized. Developed proxies were able to estimate the expected magnitudes of some of the critical performance indicators for a given set of process design parameters for CO2 and N2 injection. An inverse proxy is developed that goes beyond the capabilities of a reservoir model by providing optimized combination of design parameters for a given set of desired performance characteristics. Then, the methodology is extended to reservoirs with different characteristics by including reservoir properties in the knowledge base. Universal proxies are developed which can accurately output the performance indicators when the reservoir characteristics and design parameters are input. The hybrid neuro-genetic approach is utilized for optimization studies. This approach uses the genetic algorithm (GA) as the optimization tool. GA uses the neural-network based proxy approximator for evaluation of the specified objective function for computational efficiency. By changing the design parameters, it searches for the best design scenario that would maximize/minimize the objective function that is specified based on the nature of the problem. The reservoir model used in this study is a single-well, compositional reservoir model with a dual-porosity system. A detailed parametric study is conducted using the reservoir model to develop a better understanding of how operational parameters and reservoir conditions affect the performance of the process. Peak oil rate, discounted incremental oil production, and net present value (NPV) are used for performance evaluation. It was observed that the injected gas volume in each cycle is the most critical parameter in affecting the performance. Injecting the same volume at a higher rate for a shorter period of time is found to be more effective than injecting for a longer time at a lower rate. While soaking has little effect as compared to other design parameters, optimization of soaking would yield higher recovery and NPV. It was observed that the initial pressure/temperature of the reservoir, and therefore, the initial fractions of gas/liquid phases affect the process efficiency significantly.