Lithotype Clustering in Multidimentional Space

The focus of this work is lithotype clustering using Artificial Neural Network (ANN) for carbonate reservoirs. In contrast to terrigenous deposits, the carbonates are characterized by many parameters. It is quite a difficult problem to resolve carbonates into different lithotypes based on well data. In this work we used various data of real carbonate reservoir: well logs, core samples, petrophysical researches. To classify the carbonates several characteristics were picked out: class of carbonate (dolomites, limestone, etc.), biota, structure, recrystallization, leaching, incorporation, secondary mineral formation, type of collector, capacity. It is known that many types of data bear the curse of dimensionality. This can be mitigated by using Principal Component Analysis (PCA). PCA enables to determine the significant and uncorrelated variables. Then, the hypothesis of lithotypes discriminability on the multidimensional cross-plot should be confirmed. Finally, ANN for each characteristic was built. Laboratory results of litotype interpretation were used as training and testing data. Input neuron layer was formed by PCA output variables. Output neuron layer consisted of values of given characteristic. Lithotypes were determined by obtained characteristics and compared with laboratory results. Developed lithotype clusterization technique was successfully applied for real carbonate reservoirs