Development of proxy model for production forecast using adaptive neuro-fuzzy inference system and experimental design

Abstract

Proxy-models are computationally cheap alternative to full numerical simulation during production performance predictions. They are widely use in reservoir management to forecast production in order to assist investment decisions. However, the underperformance of many E&P projects is due to unrealistic forecast quantities arising from assumptions, human biases and reservoir modelling limitations. Hence, considerable efforts are needed to bridge gaps between forecasts and the actual production. The reservoir under study is developed with six producing wells, all deviated. The internal reservoir heterogeneity believed to have created significant fluid flow anisotropy which trapped the remaining mobile oil in the compartments poorly contacted by the current producing well spacing. Consequently, the proposed reservoir management is infill drilling. This study utilised simulation model for well selection and its optimal placement within the reservoir. Experimental design technique and adaptive neuro-fuzzy inference system (ANFIS) were integrated to develop predictive model for production forecast given new development option. Comparison of the conventional response surface (RSM) and ANFIS models was made based on their prediction performances. The ANFIS model was found to be superior.