Multi-attribute Analysis of 4D Anomalies Using Pattern Recognition Technology

The process of visual inspection of time-lapse seismic is improved considerably by analyzing multiple attributes simultaneously and by analyzing the resulting 4D anomalies in three dimensions. A pattern recognition method is used to combine complementary information from multiple attributes and detect 4D anomalies. The methodology consists of two phases; the analysis phase and recognition phase. The analysis phase aims at finding representative examples of 4D anomalies. In the recognition phase, these examples are used to train a neural network to recognize that the response of 4D anomalies differs from the background response. It is shown that the selection of examples is of crucial importance and should correspond to the objective of a time-lapse study. A comparison to conventional single attribute analysis shows the considerable reduction of non-repeatable noise. Moreover, it illustrates the potential of the method to steer the time-lapse analysis to distinguish different types of 4D anomalies.