Neural network model for the prediction of water aquifer dimensionless variables for edge- and bottom-water drive reservoirs

Accurate estimation of water influx into a petroleum reservoir is very impo rtant in many reservoir-engineering applications, such as material balance calculations, design of pressure maintenance programs, and advanced reservo ir simulation studies. These applications have relied heavily on the classi cal work of van Everdingen and Hurst for edge-water drive reservoirs and on the results presented by Coats and Allard and Chen for bottom-water drive reservoirs. However, for both types of reservoirs, the determination of the values of water influx is not a straightforward task. Table lookup and int erpolation between time entries are needed, and furthermore, for finite aqu ifers, interpolation between tables may also be required. This paper presen ts neural network (NN) models for the prediction of dimensionless water inf lux and dimensionless pressure for finite and infinite edge- and bottom-wat er drive reservoirs. Several neural network architectures using back-propag ation with momentum for error minimization were investigated to obtain the most accurate results. In order for these NN models to be applied for a wid e range of systems, dimensionless groups characterizing water influx were e mployed. The advantage of the proposed NN models is providing accurate resu lts in minimum time. Furthermore, they can be easily integrated within gene ral reservoir management programs to determine the aquifer effect on oil an d gas production.