This paper considers the application of a genetic algorithm to the steady state optimization of a serial liquid pipeline. Genetic algorithms are search procedures based upon the mechanics of natural genetics, combining a Darwinian survival-of-the-fittest with a randomized, yet structured information exchange among a population of artificial chromosomes. Computer results show surprising speed as near-optimal results are obtained after examining a small fraction of the search space. The method is ready for application to more difficult optimization problems in pipeline engineering.


Over the years, many methods have been applied to optimize both the design and operation of pipeline systems. Some methods, like dynamic - programming, require an unholy mixture of model and optimization procedure which thwarts the construction of modular programs. Other methods such as calculus-based gradient techniques require the construction or approximation of derivative information, and even then these methods can only hope to achieve local optima. As a result, there is still a need for optimization procedures which 1) are free from a particular program structure, 2) require a minimum of auxiliary information to guide the search, and 3) have a more global perspective than many of the techniques in common usage. In this paper, an algorithm with these characteristics is examined in detail. This algorithm, called a genetic algorithm (GA), is based on the mechanics of natural genetics [1]. It searches through large spaces quickly even though it only requires payoff information. Furthermore, because of the tremendous amount of processing leverage associated with genetic algorithms, the method has a much more global orientation than many methods encountered in engineering optimization practice [2]. In this paper, we first outline the pipeline problem we consider. Next, we examine the mechanics and power of effect of a simple genetic algorithm.


Suppose we have a serial products pipeline consisting of 10 pipes and 10 pump stations with 4 pumps in series within each station as depicted schematically in Figure 1. Together, these relationships specify the pressure-flow-power behavior of the serial pipeline system. For our purposes, we may think of these relationships as a black box system where we specify a set of 40 pump status variables x as input and receive a power consumption calculation as output. Actually? we also require notification of any constraint violations, but we will have more to say about that later.


To use a genetic algorithm on this or any other problem, we must adjust our-thinking to some GA differences. Genetic algorithms are different from the normal search methods encountered in engineering optimization in the following ways:

  1. GA's work with a coding of the parameter set, not the parameters themselves.

  2. GA's search from a population of points, not a single point.

  3. GA's use probabilistic transition rules, not deterministic transition rules. Genetic algorithms require the natural parameter set of the optimization problem to be coded as a finite length string.

This content is only available via PDF.
You can access this article if you purchase or spend a download.