Dynamic and recursive oil-reservoir proxy using Elman neural networks

In this work, a reservoir simulation approximation model (proxy) based on recurrent artificial neural networks is proposed. This model is intended to obtain rates of oil, gas and water production at time t+1 from the respective production rates, average pressure and water cut at t time and the well operation points to be applied in t + 1. Also, this model is able to follow the dynamics of the reservoir system applying online learning from real production observed values. Also, this model allows perform fast and accurate production forecasting for several steps using a recursive mechanism. This model will be inserted into an oil-production control tool to find the optimal operation conditions within a forecast horizon. The obtained outcomes over the approximation tests indicate the methodology is adequate to perform production forecasts.