This paper presents a new approach to well interpolation using interpolation neural networks (INETs). Traditional methods such as geostatistics have been applied to the spatial mapping of reservoirs. However, these methods are not able to make use of qualitative information, such as previously-constructed "expert knowledge" in the form of iso-porosity-contours or structural maps of sand body geometry, in a simple manner.

This paper demonstrates the usefulness of INET via a case study of a fluvial sandstone reservoir at an oilfield in the Asia-Pacific region. The proposed method is applied to porosity interpolation based on data from spatially-dispersed wells and regional geological knowledge. The results from this study show that an INET is not only able to incorporate expert advice, but also that it is easy to implement in a desktop computer or workstation. This allows an effective transfer of geological knowledge to reservoir modelling.