Well Placement Optimzation Under Geological Uncertainty

Abstract

Investments in oil and gas projects are driven by critical field development decisions including well placement, which often significantly affect the projects’ economics. Due to their typically high cost and inherently scarce data (especially in green fields, or fields in their early stage of development), managing uncertainty is critical when optimizing a field development plan. Consequently, robust field development plans require multiple geological realizations covering the range of uncertainty in reservoir properties and encompassing both multiple geological concepts and geostatistical properties distribution. On the other hand, the field development planning process involves the assessment of many development scenarios built through a combination of multiple decision parameters including, but not limited to, facility capacity, facility location, well count, well pattern, well location, well type, well trajectory, drilling schedule and operation constraints. These scenarios, when simulated over the full set of realizations, may lead to thousands of simulation runs, which are, often, practically infeasible to be conducted without an efficient automated optimization algorithm.

This work proposes a novel approach for well placement optimization under geological uncertainty. It was built on four main pillars that represent the main contributions of this work. First, we introduced a static map-based method, the Black Hole Operator (BHO), that optimizes production and injection wells’ placement based on a pre-defined well spacing (minimum distance between wells). The proposed method is systematically and thoroughly validated using a publicly available synthetic field (Olympus) that is inspired by a green oil field in the North Sea and developed for the purpose of a benchmark study for field development optimization. Results clearly illustrate the proposed method’s capability of efficiently and robustly identifying optimal well placement in comprehensive scenarios including vertical and horizontal wells in pattern and peripheral water injection schemes.

In the second contribution, we introduced a new hybrid evolutionary optimization method; the Black Hole Particle Swarm Optimization (BHPSO) for simultaneously optimizing well count, location, type, and trajectory. BHO was merged to the traditional particle swarm optimization (PSO) algorithm, where, for each particle in a BHPSO “iteration”, the location of the first producer is identified by PSO based on a net hydrocarbon thickness map. The remaining wells (producers and injectors), whose number is also potentially decided by PSO as an optimization parameter, are then automatically and optimally placed using BHO. The computational complexity of the proposed method is, thus, independent of the number of optimized wells. This drastically reduces the number of optimization parameters and, hence, the computational requirement to converge to an optimal solution. Validation results on the Olympus field show a systematically superior performance of the proposed BHPSO algorithm compared to the standard PSO.

In the third contribution, we proposed a new approach for managing uncertainty while employing the BHPSO algorithm. The statistical net hydrocarbon thickness (SNHCT) map was introduced to guide the BHPSO algorithm; and hence, pragmatically account for uncertainty in the process of well placement optimization. We optimize well placement on the realization corresponding to the minimum absolute difference between its NHCT map and the SNHCT map. The SNHCT combines the average and the P90 NHCT maps; hence, assuring that the selected sweet spots for well placement are statistically the best, with regard to the multiple subsurface realizations. The method is applied on the Olympus benchmark case and results are compared to two scenario reduction methods: RMfinder and k-means-k-medoids Clustering. Results show superior performance over the two methods in terms of optimality of well placement and the required computational load.

In the fourth contribution, we introduced, a novel machine learning based optimization algorithm for well trajectory design that achieved significant improvements in computational time compared to traditional optimization approaches. We used the Bézier curve to model the well trajectory then employed an optimization workflow to minimize the total well measured depth while honoring a dogleg severity constraint. We used the differential evolution (DE) optimizer to generate a large synthetic data that systematically, efficiently, and comprehensively cover the well trajectories. Three machine learning algorithms were then tested to train a model that predicts the well trajectory: artificial neural networks, support vector regression, and random forests. Using a machine learning model to design a well trajectory was three orders of magnitude faster than the differential evolution algorithm which, in turn, was the fastest among the different optimization algorithms that we have tested.