engleski

Review of neural network analyses performed in Croatian part of Pannonian basin (petroleum geology data)

Neural networks represent very strong tools for different prediction tasks in many sciences. Petroleum geology, and geology overall, is one of the fields where such networks can be very successful and relatively easy applied. Neural algorithms can be applied for prediction of different variables, like porosity, depth, lithology and saturation. The basic idea encompasses the correlation of several inputs and calculation of single output (predicted) value. Up to now, the neural prediction was applied at three Croatian oil and gas fields. The first analysis was done at the Okoli field (1) (prediction of facies). It is followed by porosity prediction performed at the Beničanci field (2) and finally lithology and saturation had been simulated at the Kloštar field (3). These three applications of neural networks have been performed in Miocene sediment. (1) Neural analysis performed at Okoli field in 2006 was one of the first published analyses of such type in hydrocarbon reservoir analysis in Croatia. This study is relevant for clastic facies prediction in Lower Pontian deposits of the Sava depression. Analysis is characterised by excellent correlation between predicted and true position of sandstone lithology (reservoir). On contrary, positions of predicted and true marlstones positions (in top and bottom) mostly do not correspond. The correct facies prediction probabilities are theoretically extremely high. In B-1 well (based on 3 log curves) it is minimal 78.3%, and in B-2 well (based on 7 log curves) minimal 82.1%. The Face machine is calculated relatively in the early period of network training. In B-1 well this machine is observed in 2186th iteration and in B-2 well in 7626th iteration. Such results point out that, for similar facies analyses in the Sava depression, one does not need to use such large iteration set (about 30000). Eventually, in the following neuron analyses in clastic deposits of Pannonian and Pontian ages, input dataset would need to be extended on to other well log curves. Such curves would need to well characterize lithology, porosity and saturation, like the curves of SP (spontaneous potential), CN (compensated neutron), DEN (density) and other. Presented neural technique could be useful in log curves analysis, if the Face machine would be configured with 90 % probability for true prediction. (2) At the Beničanci field (2007) the neural network was selected for handling uncertainties of porosity distribution in breccia-conglomerate carbonate reservoir of the Badenian age. The best porosity training results are obtained when all three seismic attributes (amplitude, frequency, phase) were used. The reached correlation is R2=0.987 and convergence criteria   2=0.329. These values can slightly (a few percent) differ in every new training, what is the consequence of the random sampling process in the network fitting process. The result indicates that neural network very favour the numerous inputs and it is why the meaningful variables need to be carefully selected as neural input. (3) At the Kloštar field (2007) several artificial neural networks were trained with the task of predicting lithology of Upper Pannonian deposits (“ 2nd sandstone series” ) and Lower Pontian deposits (“ 1st sandstone series” ) as well as hydrocarbon saturation within these beds. Sandstone facies represent very adequate media for statistical and neural network analysis. In the case of lithological prediction on well Klo-A and Klo-B with RBF and MLP neural networks, excellent correspondence of the true and predicted values was achieved. Prediction of hydrocarbon saturation on well Klo-B with a neural network trained on well Klo-A gave excellent corresponding between real and predicted values. Acquired results show large potential of neural networks application in petroleum geology research where they could be used as a method for acquiring quick and meaningful results from well logs or seismic data.

neural network; prediction lithology; porosity; saturation; Drava depression; Sava depression; Pannonian basin; Croatia

nije evidentirano