Generating Synthetic Well Logs by Artificial Neural Networks (ANN) Using MISO-ARMAX Model in Cupiagua Field

Abstract

Abstract Well logs acquired directly in field have turned out to be one of the most key engineering elements to evaluate hydrocarbon formations. Nevertheless, the lack of information, some technical troubles related to the unfolding of tools, the operational states of the well and many other reasons may sharply limit the carrying out of an optimal formation characterization methodology along the entire productive or injective lifespan of a reservoir. Nowadays, artificial neural networks (ANN) are one of the strongest tools to supply such missing information in order to generate synthetic logs. In this paper, we explain the putting into practice of an ANN methodology with the aim of provide useful input information in geomechanical modeling for the hydraulic fracturing simulator GIGAFRAC. More explicitly, the purpose of the schemes presented here is to provide transit-time curves for primary or compressional waves (DtP) and secondary or shear waves (DtS), based on full information measurements of Gamma Ray, Neutron-Porosity, Density, DtP and DtS logs; for some wells in the Cupiagua field located in Colombian Foothills, which break through some geologic formations such as Mirador, Barco, Guadalupe, and Los Cuevos. A noteworthy amount of considerations were taken into account to ensure the success of the ANN estimation phases. A strong focus is done regarding to filtration and quality control of the input information to the network, relating to the control mechanism of outliers, as well as the splitting-up of logs in zones by using a geological criteria and spreading of data information in computationally convenient vectorial and matricial arrangements. Finally, good adjustments were obtained throughout the validation phases and they all were considered as successful outcomes, together with training phase and subsequent use of the same for estimating DtP and DtS curves.