Constrained Gas Network Pipe Sizing with Genetic Algorithms

The application of a genetic algorithm to an industrial pipe-sizing problem is presented. The optimum sizes for the pipes in a given network with specified supply and demand requirements must be determined, subject to two additional constraints. The problem was used as a test case for the evolutionary computing language RPL2. The genetic approach is shown to produce better results than the existing industrial heuristic at the expense of longer run times. 1 Problem Specification It is a frequent criticism of genetic algorithms and evolutionary computation more generally that published problems are usually contrived and unconstrained. (There are of course notable exceptions, for instance the work of Goldberg (1989) on gas network compressor optimisation.) This paper demonstrates how a relatively straightforward implementation of a genetic algorithm can find significantly better solutions than the standard heuristics actually used for a real, constrained pipe-sizing problem to offer actual cost reductions of about 4%. The design of a gas network—e.g. to supply a new housing development—involves defining the layout of the network and, having done this, choosing the types of pipe to be laid. The layout is generally determined by such considerations as the routes of roads but the selection of the pipe types is tackled as a constrained optimisation problem. The important constraints on any design are that: the pipes selected should allow the customer demands to be met at or above a ‘minimum design pressure’ each pipe (other than those incident to a source) should have at least one upstream pipe of the same or greater diameter. Pipes are produced in a range of discrete diameters and in a number of materials, and for a given material the cost per unit length of pipe is an increasing function of diameter.