No description available.

No description available.

No description available.

No description available.

No description available.

No description available.

No description available.

Shale gas development is transforming the energy landscape in the United States. Advances in production technologies, notably the dual application of horizontal drilling and hydraulic fracturing, allow the extraction of vast deposits of trapped natural gas that, until recently, were uneconomic to produce. The objective of this work is to develop mixed-integer programming models to support upstream operators in making faster and better decisions that ensure low-cost and responsible natural gas production from shale formations. We propose a multiperiod mixed-integer nonlinear programming (MINLP) model along with a tailored solution strategy for strategic, quality-sensitive shale gas development planning. The presented model coordinates planning and design decisions to maximize the net present value of a field-wide development project. By performing a lookback analysis based on data from a shale gas producer in the Appalachian Basin, we find that return-to-pad operations are the key to cost-effective shale gas development strategies. We address impaired water management challenges in active development areas through a multiperiod mixed-integer linear programming (MILP) model. This model is designed to schedule the sequence of fracturing jobs and coordinate impaired- and freshwater deliveries to minimize water management expenses, while simultaneously maximizing revenues from gas sales. Based on the results of a real-world case study, we conclude that rigorous optimization can support upstream operators in cost-effectively reducing freshwater consumption significantly, while also achieving effective impaired water disposal rates of less than one percent. We also propose a multiperiod MINLP model and a tailor-designed solution strategy for line pressure optimization in shale gas gathering systems. The presented model determines when prospective wells should be turned in-line, and how the pressure profile within a gathering network needs to be managed to maximize the net present value of a development project. We find that backoff effects associated with turn-in line operations can be mitigated through preventive line pressure manipulations. Finally, we develop deterministic and stochastic MILP models for refracturing planning. These models are designed to determine whether or not a shale well should be restimulated, and when exactly to refracture it. The stochastic refracturing planning model explicitly considers exogenous price forecast uncertainty and endogenous well performance uncertainty. Our results suggest that refracturing is a promising strategy for combatting the characteristically steep decline curves of shale gas wells.

Linear programming (LP) models that contain a (substantial) network structure frequently arise in many real life applications. In this thesis, we investigate two main questions; i) how an embedded network structure can be detected, ii) how the network structure can be exploited to create improved sparse simplex solution algorithms. In order to extract an embedded pure network structure from a general LP problem we develop two new heuristics. The first heuristic is an alternative multi-stage generalised upper bounds (GUB) based approach which finds as many GUB subsets as possible. In order to identify a GUB subset two different approaches are introduced; the first is based on the notion of Markowitz merit count and the second exploits an independent set in the corresponding graph. The second heuristic is based on the generalised signed graph of the coefficient matrix. This heuristic determines whether the given LP problem is an entirely pure network; this is in contrast to all previously known heuristics. Using generalised signed graphs, we prove that the problem of detecting the maximum size embedded network structure within an LP problem is NP-hard. The two detection algorithms perform very well computationally and make positive contributions to the known body of results for the embedded network detection. For computational solution a decomposition based approach is presented which solves a network problem with side constraints. In this approach, the original coefficient matrix is partitioned into the network and the non-network parts. For the partitioned problem, we investigate two alternative decomposition techniques namely, Lagrangean relaxation and Benders decomposition. Active variables identified by these procedures are then used to create an advanced basis for the original problem. The computational results of applying these techniques to a selection of Netlib models are encouraging. The development and computational investigation of this solution algorithm constitute further contribution made by the research reported in this thesis.

No description available.