Well Placement Optimization

Challenge

Efficient well placement in complex reservoirs is critical for maximizing hydrocarbon recovery while minimizing drilling risks and operational costs. Traditional geosteering relies heavily on human interpretation and may not fully exploit real-time measurement-while-drilling (MWD) and logging-while-drilling (LWD) data. We have access to historical data from a well drilled in the North Sea, including the measurements recorded and the steering decisions made by human operators. This provides a unique opportunity to analyze how humans made geosteering decisions and to compare those decisions with outcomes generated by algorithmic optimization. Key challenges include uncertainties in subsurface models, layer dips, and drilling dynamics, which can lead to suboptimal well trajectories.

Innovation

Develop a real-time, multi-objective geosteering workflow using well data to determine the optimal next-step trajectory ahead of the bit, test multiple optimization algorithms, identify key objective function parameters to improve well placement and reduce uncertainty and finally compare results with the original drilling decisions for a well drilled in the North Sea.

Value

Evaluation of multiple multi-objective (drilling dynamic + reservoir model) optimization algorithms  for well placement and next-step trajectory decisions using a well data from the North sea, identification of key objective function parameters and terms that most influence trajectory optimization, and comparison of algorithmic decisions with historical human steering decisions for this specific wellbore.

Status

We have developed several models capable of generating both reservoir parameters (such as electrical resistivity and flow parameters) and  drilling dynamics (weight-on-bit, torque, ROP, tortuosity). The next step is to integrate these parameters into a multi-objective function that can be used to evaluate their relative importance for well placement decisions. We plan to implement and test different optimization strategies, starting with classical algorithms (e.g., evolutionary approaches) and advancing to deep learning (DL)–based methods.