Porosity estimation improvement by averaging technique from well log in Balal oil field

Estimation of porosity in hydrocarbon reservoirs is essential for planning production operations. Lateral variations of porosity cannot easily bedelineated from measurements made at sparsely located wells(Soubotcheva, ٢٠٠٦; Hampson and others, ٢٠٠١; Soto, ١٩٩٨). So, the integration of ٣D seismic data with petrophysical measurements cansignificantly improves the spatial distribution of porosity. Despite sparse well data, ٣D seismic data provide a dense and regular areal sampling ofthe acoustic properties of the producing reservoirs. After processing of ٣D data, the lateral variations of seismic amplitudes can be transformed into impedances by integrating it from the well and geological data, which in turn are indirectly related to porosity (Pramanik and others, ٢٠٠٤; Todorov, ٢٠٠٠; Angelier and Carpi, ١٩٨٢; de Buyl and others, ١٩٨٦).Artificial neural networks (ANNs) are very suitable technique in softcomputing for signal processing. According to a set of multivariate input and target measurements, ANNs can learn and extract their complex nonlinearrelationships. The relationships can be applied to estimate the target variables when the actual measurements are not available (Wong and others, ٢٠٠٢; Ronen and others, ١٩٩٤). Previous studies by this method have shown good results in field applications, compared to the wellestablishedmethods such as multiple linear regression and discriminant analysis. So, this method has been used in the paper (Al-Bulushi andothers, ٢٠١٠; Wong and others, ٢٠٠٧; Wong and others, ٢٠٠٢). Because frequencies of well logs and attributes aren’t identical, onlysamples of attributes that is correlated temporally with samples of target log are inserted to calculations. Multivariate regression method had beendeveloped by Hampson to solve this problem that convolution filters are used instead of single points (Hampson and others, ٢٠٠٠; Russell andothers, ١٩٩٧; Russell, B. H., ٢٠٠٤). This method is equivalent with creating a set of new attributes that in comparison with main attributes had beenshifted temporally. This time shifts are coincident with convolution filters. But many samples, on the different attributes, aren’t inserted intoestimation process because of frequencies distinction and in fact these samples don’t have any role in estimation. It can be inserted average oflogs instead of porosity logs because of the studied horizon has homogeneity petrophysically and in reservoir properties and there is littlechanges in porosity. So with averaging from logs and attributes in the horizon, both the problem of distinct frequencies is solved and lower errorare obtained. So, main goal of this paper is studying of results obtained from porosity estimation by using artificial neural network before and after averaging from logs and seismic attributes in studied reservoir horizon. To achieve the defined goal, one of the southern Iranian oil fields is selected.