Neural Network as an Alternative to Rock Physics Modeling in Time-Lapse Seismic Reservoir Monitoring

Abstract

Abstract In this paper we present a neural network based approach to rock-physics modeling in the context of time-lapse seismic analysis. We aim to find the optimum relationship between pressure and saturation one hand and seismic velocity on the other. Such a relationship is essential in time-lapse interpretation, especially when involving reservoir parameter inversion and forward modeling. The training set for our network is constructed from reservoir simulator data and inverted seismic velocities. To test the proposed methodology, it has been applied to Statfjord field data. The results are compared to the current Statfjord rock-physics model, which considers both saturation and pressure. Our approach can be regarded as an alternative method to study rock physics behaviour. It creates an opportunity to study the reservoir behaviour at the seismic level, hence at a low seismic frequency. Ultimately time-lapse seismic interpretation can be improved if we obtain a better understanding of the relationship between rock properties and seismic properties. Introduction From the 1940s to about a decade ago, the main use of reflection seismic data in the oil industry has been structural subsurface mapping. In the 1980s, the traditional 2D survey evolved into 3D seismic survey due to ongoing developments of modern computing techniques. Over the last 10 years, 3D surveys are more and more used for reservoir characterization purposes to identify static properties, e.g. porosity. Currently, time-lapse or 4D seismic is becoming a regularly used technique. Time-lapse seismic allows identification of dynamic reservoir behavior by interpreting the observed differences over time, as induced by production. Understanding the relation between seismic properties of reservoir rock and the reservoir state is a crucial aspects of time-lapse seismic reservoir monitoring. After all, we want to interpret the observed time-lapse seismic changes in terms of reservoir engineering properties, e.g. fluid content. The enduser of time-lapse seismic (e.g. the reservoir engineer) cannot use the black-and-white loops or seismic attributes directly, but requires a translation to reservoir parameters. These inverted reservoir parameters are used to verify the reliability of the reservoir simulation model and, if necessary, to update the reservoir simulation model. During production of an oil/gas reservoir, the saturation, pressure, and temperature can change, thereby inducing a certain change in the seismic signal over time. For interpret purposes one has to assume repeatability, i.e. the observed seismic signal can be attributed to the change in reservoir state and not to acquisition and processing artifacts. The inversion of the time-lapse signal to these three dynamic reservoir properties requires knowledge of the exact impact on seismic of each parameter. Rock physical models describe the behaviour of the rock excited by a seismic wave based on observations and lab experiments. However, the actual behaviour of a fluid-filled rock, excited by a seismic wave, is extremely complex and not well understood yet. Generally, rock physics models describe the effective rock behaviour and they are not always capable of accurately describing the impact of either saturation, pressure or temperature.