Development of an artificial neural network model to predict future oil production rates in a sandstone reservoir under gas injection

In petroleum industry there have always been an urge to improve hydrocarbon recovery, especially in the depleted fields in which the reservoir pressure is not high enough to satisfy the economic goals of theowners. Low reservoir pressure, unrecoverable oil trapped in the reservoir, etc. are amongst many reasonsthat make the use of recovery methods other than the primary techniques inevitable. Enhanced oil recovery (EOR) is a generic term for those techniques which are employed to increase therecoverable oil from such reservoirs. Gas injection has proved to be one the most effective and as a result most common EOR methods in thepetroleum industry. This method helps maintain the reservoir pressure and also would mix with thetrapped oil and lower its viscosity and push it towards the production wells. As the gas injection project like any other EOR method is going to be rather expensive to be applied in a reservoir, prediction of the field responses to the injection and prediction of the production rate resulting from the injection process isof great importance as these will lead to an optimized injection scheme. In this paper artificial neural networks (ANN) are used to find a meaningful relation between different properties and variables of an injection-production system simulated by Eclipse reservoir simulator basedon a real reservoir in North Sea to realize if ANN is capable of the prediction of the injection-production data.

Keywords:

Gas Injection , Artificial Neural Networks , MLP Neural Networks , ECLIPSE , Reservoir Simulation , Data Normalization , Noise Addition