Advanced well structures: An artificial intelligence approach to field deployment and performance prediction

Advanced well structures present themselves as a veritable tool tor exploiting unconventional resources. These well systems reduce surface footprint of drilling operations whilst enhancing production by increasing contact area with the reservoir. Traditional methods for well performance analyses barely capture the complex interaction between these well structures and the reservoir. Therefore, this paper discusses the application of artificial neural network models to forecast production from advanced well structures, and well designs for field deployment of these well structures in unconventional multi-phase reservoirs. Two forward-acting models and two inverse-acting models were developed. The forward-acting models predict oil, water and gas production profiles (and bottom-hole pressure profile for flow rate-specified wellbore conditions) for any given well design with parameters within the range of values used in training the model. The inverse-acting models generate well designs that can meet a desired oil cumulative production profile. Gas and water cumulative production, and/or bottom-hole pressure profiles are also predicted by the inverse models. For each category, one model was developed for constant pressure wellbore conditions, while the other model was developed for flow rate-specified wellbore conditions. Predicted well designs were validated using high fidelity simulators. Synthetic field data was used for training, which were drawn from a 574-acre, isotropic and homogenous, naturally fractured reservoir system that represents average reservoir characteristics. Matrix permeability was 0.01 mD while fracture permeability was 5.0 mD. This reservoir system was utilized for each model. Well design parameters include the length and location of the horizontal mainbore; number of side laterals; length, spacing, and direction of each lateral. The developed models showed good performances with minimal prediction errors. These results are promising, lending credence to the application of artificial intelligence for even more complex reservoir systems. The observed results should also boost confidence and interest in the use of advanced well structures for field applications.