Abstract

Petro-Physical parameters obtained from the well logging data is crucial in decision making for reserve estimation. Tool response is an important criteria in the acquisition of good quality data. Malfunctioning of logging tool leads to data loss while recording the data. To overcome this problem, a methodology has been developed to generate the missing log from the other nearby wells log data. In one well of CBM block located in Bokaro field the density log data could not be recorded due to tool malfunctioning. The developed methodology is followed to predict the missing data using seven nearby well data. The density data is selected as the target log with the training of external attributes. Single attributes analysis is carried out and number of regression equations are generated. The attribute with minimum error and maximum correlation relationship is selected as input to initiate the multi-attribute analysis. Multiattribute analysis are carried out by creating a list of transform by stepwise regression between a subset of the attributes and the target log values from single attribute. Proper gain and lag are applied by Probabilistic Neural Networks (PNNs) using the concept of distance in attribute space from known to unknown point. Attribute with minimum value in training and validation error is selected to obtain the relationship between the predicted and observed density data to generate the final data. The predicted density data is validated with nearby well data. The result obtained using multi attribute and PNN has correlation 149.2 and error 0.91 compared to single attribute correlation 0.80 and error 214.5. This methodology of artificial intelligence helps to predict the missing log with accuracy and to obtain sub-surface information.