Forecasting Well Performance in a Discontinuous Tight Oil Reservoir Using Artificial Neural Networks

Abstract

Abstract Improving the economics of the production and development of an unconventional reservoir system is a key to meeting increased demand for hydrocarbons in the near future. In general, reservoir development is vastly assisted by using hard-computing models to evaluate the potential of the formation. These models have been used to identify infill drilling locations and forecast production. However, preparing the simulation models for discontinuous tight oil reservoir systems poses a challenge with hard-computing protocols. This paper discusses a methodology developed to depict the production characteristics of a reservoir via the geological properties of the reservoir. The methodology discussed in the paper is time efficient and is proven to generate effective results. The methodology discussed in the paper utilizes Artificial Neural Networks (ANN) to map the existing complex relationships between seismic data, well logs, completion parameters and production characteristics. ANNs developed in this work are used to forecast oil, water and gas cumulative production for a two year period. The results obtained are also extended to identify potential infill drilling locations. This work enables the practicing engineer and the geoscientist to analyze an entire reservoir in a time efficient manner. The workflow is demonstrated on a discontinuous tight oil reservoir located in West Texas. The results discussed in the paper show the robust nature of the methodology. The workflow also helps in improving the resolution of the production surfaces which help in identifying productive, yet undrilled, locations in the reservoir. The production surface for the entire field is forecasted within a one minute time frame (~6600 locations). The method developed will help in avoiding low producing wells prior to drilling, and thus, is expected to help in the economic development of complex tight oil reservoirs.