Prediction of lateral variations in reservoir properties throughout an interpreted seismic horizon using an artificial neural network

Abstract

Successful use of an artificial neural network is shown to predict lateral variations of seismic velocity, density, thickness, and gamma rays associated with sand dune reservoirs identified on a previously interpreted seismic horizon. The work is presented in two main sections. Section one is a feasibility analysis based on synthetic data. A known geologic model is used, performed by pseudowells, in which lateral variations in seismic velocity, density, and gamma ray values are related to the dunes. The synthetic seismic model and the attributes derived are used as training input in the neural network. Section two is a real case example where the methodology is applied to a real seismic data set. Results indicate that using a set of data and attributes restricted to a time interval corresponding to a previously interpreted seismic horizon is more efficient than using a whole data cube, involving a very large volume of data.