Porosity estimation of pore volume in one of the Persian Gulf's hydrocarbon fields use multi-attribute analyses

Integration of 3D seismic data with petrophysical measurements gives a better vision to reservoir characterization. The integration of well-logs and seismic data has been a consistent aim of geoscientists which become increasingly important and successful. In recent years، because of the shift from exploration to development of existing fields with a large number of wells penetrating them، improving reservoir study has been the most important pre-drilling activity. One type of integration is forward modelling of synthetic seismic data from the logs. A second type of integration is inverse modelling of the logs from the seismic data. It is called seismic inversion. Seismic inversion، in geophysics، is the process of transforming seismic reflection data into a quantitative rock-property description of the reservoir. Another method is to estimate the log properties by seismic attributes. In this study، linear multiattribute transform، and non-linear multi-attribute transform were used for predicting porosity in one of the Iranian hydrocarbon fields. The analysis data consisted of the target log (in this study، the porosity log) from wells tied with 3D seismic volume. From the 3D seismic volume، a series of sample-based attributes was calculated. The objective was to derive a multi-attribute transform، which was a linear or nonlinear transform between a subset of attributes and target log values. The selected subset was determined by a process of forward stepwise regression، which derived increasingly larger subsets of attributes. In the linear mode، the transform consisted of a series of weights derived by least-squares minimization. These weights are coefficients of the selected attributes in a linear multi-attribute transform. In the nonlinear mode، a neural network was trained using the selected attributes as the input. Two methods of neural network used in this study include probabilistic neural network and multi-layer feedforward network. The basic idea behind the general regression probabilistic neural network is to use a set of one or more measured values، called independent variables، to predict the value of a single dependent variable. The multi-layer feed-forward network method consists of a set of neurons، arranged into two or more layers. There is always an input layer and an output layer، each containing at least one neuron. Between them، there are one or more ‘hidden’ layers. The neurons are connected in the following fashion: inputs to neurons in each layer come from outputs of the previous layer، and outputs from these neurons are passed to neurons in the next layer، and each connection represents a weight. To estimate the reliability of the derived multi-attribute transform، cross-validation was used. In this process، each well was systematically removed from the training set، and the transform was rederived from the remaining wells. Then، the prediction error was calculated for the hidden well. The validation error، which is the average error for all hidden wells، was used as a measure of the likely prediction error when the transform was applied to the seismic volume. There was a continuous improvement in predictive power as it was progressed from a single-attribute regression to a linear multi-attribute prediction. This improvement was evident not only in the training data but، more importantly، in the validation data. In addition، the neural network did not show a significant increase in resolution over that from the linear regression. As a conclusion، the best result of porosity estimation in this field was provided by the linear multi-attribute transform.