Well Log Prediction via Machine Learning and Joint Inversion

The patent creates virtual copies of triple combo logs (gamma ray, resistivity, and density neutron logs) using machine learning models trained on data from offset wells. These synthesized logs replicate the information content of actual wireline and advanced LWD logs without requiring the physical logging tools, thereby reducing device complexity while maintaining measurement precision for formation evaluation.

The machine learning models are pre-trained on comprehensive datasets from multiple offset wells before the actual drilling operation. This preliminary action allows the models to predict triple combo log values in real-time during drilling without requiring actual wireline or advanced LWD logging operations, thus maintaining information completeness while avoiding the time loss associated with these lengthy logging processes.

The patent combines multiple data sources including MWD measurements (gamma ray, ROP, WOB, torque, RPM, differential pressure) with historical data from offset wells that have complete triple combo logs. The machine learning model integrates these diverse datasets to train predictions that are specifically adapted to unconventional wells while maintaining measurement precision through the combined information from both limited real-time data and comprehensive historical data.