Liquid pipelines can be very expensive to operate because of energy costs in the pumps and the expense of DRA injection.
Optimizing these pipelines presents an interesting technical challenge because of the large discrete components in the formulation: Pump unit selection and Station Bypass selection. Pump unit selection involves choosing not only the best pumps to run at a particular station to match current conditions, but also choosing system-wide pressure distribution so that the various pumping stations can support each other in collectively hitting the " sweet spots" of as many stations as possible. Choices in unit selection at one end of a pipeline can affect choices all the way at the other end, and vice versa. Bypass selection adds an even larger discrete component to the problem since running pump stations typically severely degrade drag reducing agent (DRA) in the fluid, making expensive re-injection necessary. If one or more stations can be bypassed at proper locations and in conjunction with the pump unit-selection system-wide optimization, the cost of re-injecting expensive DRA downstream of that station can be reduced.
We consider several algorithms for pump selection optimization. Exhaustive search through the possibilities is trivial to implement but has the obvious problem of exponential growth in computation required. Branch and bound algorithms can reduce the computation needed but still have issues. A hierarchical algorithm which creates local optimal composites of individual pump stations, and performs a polynomial-time variant of Dynamic Programming to solve the system-wide optimization, has proven very useful.
The pump selection optimization methods in this paper consider only operation at a single specified time point, with known flow rate and specified commanded concentrations of DRA by pipe and by batch. The benefits of this approach to pump selection are magnified if this is included as part of a larger optimization problem which manipulates both flow rates and DRA concentrations as well. Methodologies for the larger problem are beyond the scope of this presentation.
