Application of pattern recognition method in oil and gas prospection in cenomanian and malmian deposits of Nida synclinorium
More details
Hide details
Gospodarka Surowcami Mineralnymi – Mineral Resources Management 1999;15(Zeszyt specjalny 1):45–68
Sub-Miocene basement of the Carpathian foredeep contains, most probably , unrecognized areas of potential hydrocarbon deposits, occurring in analogous conditions as those already discovered. Particularly rich data come from the area, being the prolongation of Nida synclinorium and refers to Cenomanian and Malmian deposit (fig. 1). This area was intensely explored what resulted in the discovery of several deposits (figs. 1, 2). Usually they occur in the zone where Cenomanian sandstones are wedging out (trap of 1st type), as well as in weathered, porous, karstified Malmian carbonate rocks (trap of 2nd type), also in areas where Cenomanian rocks do not occur (fig. 1). Rarely, hydrocarbon deposits appear in planar domes within the top of thick Cenomanian sandstone sequences (trap of the 3rd type e.g. Rylowa deposit). All these traps are sealed by hardly permeable Upper Cretaceous or Miocene deposits. There are credible premises indicating that in some boreholes, due to imperfect drilling technology, the presence of economic hydrocarbon accumulation could be overlooked and, thus, they were classified as barren. Moreover, vast areas between almost 700 boreholes may also contain undiscovered deposits. As follows from above data, further exploration works are needed, using different methods. As far as mathematical methods are concerned, the most useful can be the pattern recognition methods (PRM), applied jointly with geostatistical ones. The experiments on application of these methods in the area studied is the aim of investigation presented in this paper. In the analyzed case, the application of PRM consist in the calculation of probability of membership of each borehole studied to one of two target classes: to class of wells with economic hydrocarbon accumulation (R) or to those barren (W) i.e. to the computation of P[R] and P[W]. This procedure should be preceded by calculation of probability of membership of each borehole to one of two predictor classes: R* and W*, connected with appropriate target classes R and W. The predictor is the mathematical representation of a given borehole. Predictors are multivariate vectors of features, the values of which are measured in each borehole. The predictor class R* (or W*) is formed by vectors of features measured in boreholes selected from class R (and respectively W) and assented as their patterns. In the analyzed case, pattern boreholes from each deposit were selected for the predictor class R*, more or less proportionally to the number of wells in each deposit (tab. 2). Pattern boreholes for predictor class W* were selected systematically from the whole surface of the areas, situated outside deposits. 16 features were selected from the set of 27 measured (or computed) for the boreholes, the values of which were obtained in the most credible manner (tab. 1). These features were characterizing the depth and thickness of Cenomanian deposits (numbers: 1, 2, 3), their porosity and permeability (numbers: 4, 6), mineralization and potentiometric heads of groundwaters (numbers: 7-10, 17), as weIl as gradients of these teatures (numbers: 22, 23, 25-27). The examination of informativity of these features by means of Puri-Sen-Tamura test of comparison of multivariate means has shown them to form one signiticantly informative combination (fig. 3). The method of potential function in probabilistic version was applied as that of PRM. Consequently, each borehole (pattern and nonpattern, altogether 691) was characterized by one probability value of membership to predictor class R*. The application of block kriging was necessary for the production of maps of the above probability for the whole area. Preliminary geostatistical studies aIlowed to apply the compound variability model to kriging computations (figs. 4, 5). The interpolation of P[R*] probabilities and kriging standard deviations reIated with them are presented in maps (figs. 6, 7). The elaboration of resulting map needs the computation of conditional probabilities (P[RIR*], P[WIR*], P[RIW*], P[WIW*]) (cf. tab. 3). Moreover, the corrections ER and EW proposed by the present authors should be taken into consideration, according to the formula: P[R] = P[R*] (P[RIR*] + ER) + (1 - P[R*]) (P[RIW*] + Ew) where: ER = P[WIR*] P[R'IW] P[R*IR] / P[R*IW] and Ew = P(WIW*] P(R'IW] P(W*IR] / P(W*IW]. These corrections take into account the effect of faulty identification of boreholes (i.e. \delta error) during drilling on the value of the membership probability of a given well to class R. The result of such erroneous identitication is that within the class W there can be incorporated the boreholes forming subclass R' i.e. those, penetrating existing hydrocarbon deposits (fig. 12). The value of this error, expressed as P(R'IW] probability, can be estimated but a priori. The effect of the error in question on prognosis is, in fact, small (tab. 4) but by taking into account the proposed corrections we are increasing the adequacy of recognition. The probability of membership of each point of the area to target class R, is presented in isoline map (cf. fig. 8). All the known deposits considered in our experiment are Iocated in this map within the areas showing the probability higher then 0.5. Moreover, new potential areas of occurrence of deposits were indicated as isolated maxima showing probability more then 0.5. They are representing first of all the traps of 1st type (arrow numbers 1, 3, 5), as well as of 2nd one (arrow numbers 2, 4, 7). Only in one case they represent also 3rd type (arrow number 6). Consequently, there were indicated first of all the regions, characterized by small thickness of Cenomanian deposits, similarly as the vast majority of known deposits that have delivered the patterns. Therefore, other potential deposits, occurring in traps of 3rd type i.e. not related with zones where Cenomanian deposits are thin or lacking, could not be discovered. In order to explain this situation, the second experiment was carried out, in which thickness and the features correlated with it: porosity, permeability and thickness gradient of Cenomanian deposits were eliminated. The result of computation, like in previous experiment, are presented in appropriate maps (figs. 9, 10, 11). Final results of recognition have revealed almost aIl the above indicated new potential zones of hydrocarbon accumulations (excluding maximum no 6) and, moreover two successive, just representing the 3rd type of traps (fig. 11, numbers 8, 9). One of the Iast zones (no 8) showing high probability of occurrence of deposit, amounting to 0.65 has appeared SW of Rylowa deposit. Irrespective of the studies presented here, the borehole Rajsko-1 has, in fact, penetrated a deposit in that zone. This is an important confirmation of reality of results obtained by means of the pattern recognition method.
Rozpoznawanie obrazów w prospekcji stref naftowych w cenomanie i malmie synklinorium Nidy
złoża węglowodorów, rozpoznawanie obrazów, geostatystyka, predykcja
Autorzy zastosowali metodę funkcji potencjalnych (metoda rozpoznawania obrazów) w połączeniu z metodami geostatystycznymi do wskazania nowych potencjalnych obszarów akumulacji węglowodorów w mezozoicznym piętrze strukturalnym podłoża zapadliska przedkarpackiego, w jego części obejmującej S przedłużenie synklinorium nidzińskiego. Obiektami badań były otwory wiertnicze opisane zestawem 16 cech informatywnych. Cechy te charakteryzują głębokość i miąższość osadów cenomanu, ich porowatość i przepuszczalność, mineralizację i napory potencjometryczne wód wgłębnych, jak również gradienty tych cech. 16-wymiarowa informacja o każdym otworze wiertniczym została ostatecznie przetworzona w prawdopodobieństwo, że otwór odwiercony w dowolnym punkcie badanego obszaru nawierci złoże. W efekcie wskazano 9 rejonów o podwyższonym prawdopodobieństwie odkrycia złoża.