Unsupervised Machine Learning Assisted Hydrogeophysical Borehole Logging Inversion for Robust Aquifer Characterization

This research proposes an integrated algorithm that uses an unsupervised machine learning technique, specifically the new K-mean clustering, for automatic aquifer characterization using hydrogeophysical borehole logging data. The MFV-cluster algorithm was employed to determine layer boundaries and petrophysical parameters automatically. The viability of the suggested process was evaluated using synthetic and field data, and it was found to be effective in distinguishing between various forms and providing a preliminary estimate for layer thicknesses. The integration between the new cluster technique and interval inversion can help with the automatic detection of both the geometrical and petrophysical parameters. The field data used in the study showed a shaly sand pattern response. The MFV clustering technique was applied to this field data and was able to distinguish between various forms and provide a preliminary estimate for layer thicknesses. The results of the statistical evaluation of synthetic data contaminated by 30 percent of outliers prove a high dependency on the initial location of the centroid. The interval inversion approach enhances the number of inverted data points by representing the petrophysical parameters as a continuous function.