We examine a number of techniques in current use in pipeline simulation for selecting which units to operate within a station and how to operate the selected units. By focusing our attention on a restricted class of nonlinear compressor models (a class that is nevertheless much more realistic than the linear models often used), we are able to compute exact solutions to the station optimization problem, and therefore have a baseline for assessing the accuracy of existing simplified methods. We present circumstances where existing heuristics come arbitrarily close to the optimal solution, as well as situations where they can be 30 % or more sub-optimal. We next examine algorithm computational speed, particularly with regard to how algorithms scale with respect to number of units within a station. By implementing the heuristics using a branch-and-bound algorithm rather than the standard algorithm, speedups of orders of magnitude can be obtained for large stations. This speedup is sufficient to allow heuristic station optimization to be used within larger pipeline simulations which invoke embedded station simulations thousands or millions of times. Finally, we present a hybrid mixed-integer-nonlinear programming method which is capable of efficiently computing exact solutions to the restricted class of compressor models presented above, and discuss straightforward extensions of this algorithm to general compressor models. In conclusion, we attempt to place station optimization in context with regard to simulation in general, appropriate objective functions, and current and future problem formulations and algorithms.
A principal component of any gas transmission system is the compressor station. A given system may have anywhere from a few stations up to well over 50. These stations add enough energy to the gas to overcome frictional losses and to maintain required delivery pressures and flows. Most compression is powered by natural gas taken directly from the pipeline, but electric powered compressors are an increasing phenomena. Fuel and power costs for compressor operation approach a staggering half billion dollars per year in the United States alone.' Every 1% in fuel savings that can be achieved therefore represents up to 5 million dollars per year in economic benefits. In this paper we examine some aspects of compressor station simulation and optimization. In particular, we examine some of the simplifications that are sometimes made in addressing this problem, and attempt to develop a broad picture of the strengths and weaknesses of various methodologies.
In Figure (1) we see a parallel unit compressor station in schematic form. Gas enters the station at a suction pressure Ps and is discharged at pressure PO, with total flow through the station designated as Qtotd. Flow through any given unit (or path) is denoted Q; for each of the N paths. Each unit may either be operating or shut down.