Application of fuzzy logic and neural networks for porosity analysis using well log data: an example from the Chanda Oil Field, Northwest Pakistan

Fuzzy logic (FL) and neural network (NNs) methods are commonly applied in a variety of areas in the petroleum industry. The area of hydrocarbon exploration has seen the greatest advancement of the soft-computing technologies including FL, and NNs. In this study, FL and NNs methods have been applied to log data from the Chanda Oil Field, northwest Pakistan, for porosity (PHIT) prediction. The input dataset for the study included four known logs, gamma ray (GR), neutron porosity (NPHI), density (RHOB), and sonic (DT) of two wells drilled in the Chanda Oil Field. For FL model, ten numbers of bins were selected. The closeness of fit (Cfit) curves were calculated considering the most, and second most likely curves. The weighted average final probability Pi, or the most likely solution, was also computed. The curve histogram distribution and the set of curve bin distribution cross plots were generated using the fuzzy model. In the FL model, the Gaussian membership function was the best fit for the well log data analyzed. FL models show Cfit fall in the range of 92–100% for Chanda-1 and Chanda Deep-1 with standard deviations 1.268 and 1.396, respectively. NNs models were generated for Chanda-1 and Chanda Deep-1 in the Datta Formation reservoir interval. The NNs model was trained using back propagation (BP) algorithm. NNs model reveals the Cfit_nn in the range of 85–100% for two wells with standard deviations 0.012 and 0.025. The results reveal a very good match between the log data and predicted modeling analyses using FL and NNs methods. These techniques can be applied to reduce uncertainty in determining the PHIT in wells. For comparison, the multiple linear regression (MLR) analysis was performed using the same log dataset of two studied wells. The coefficient of determination (R²) derived from the FL model (PHIT_ml) were 0.5727 and 0.7988, and 0.6256 and 0.8527 for NNs model (PHIT_nn) in two wells. In comparison, PHIT curve values for MLR (PHIT_mlr) were 0.512 and 0.338, The high R² values indicate FL and NNs as reliable techniques for PHIT prediction compared to MLR method. The application of FL and NNs methods to well data indicates that these two methods can better determine the PHIT with an accuracy that rivals that of other methods, such as those based on statistics such as MLR. The corresponding correlation was obtained through a comparison of synthesized log values with real log values. The comparisons between the measured and predicted parameters using the two different methods FL and NNs indicated that both were successful in synthesizing PHIT logs. This paper indicates that for studied wells in the Chanda Oil Field, both FL and NNs are reliable, giving a realistic match for real and synthesized PHIT curve using a combination of GR, RHOB, NPHI and DT logs.