| Operating a gas transmission pipeline is a challenging task which has relied upon the savvy of experienced operators for proper control. Computers have played a growing role in communications, equipment control, simulation, and optimization; however, computers have not directly aided in pipeline decision making.; In this dissertation, two techniques connected with genetics and artificial intelligence are applied to gas transmission pipeline control to approach the robustness--the efficiency and breadth of capability--of the human gas dispatcher.; First, a genetic algorithm (GA) is developed to optimize two gas pipeline problems: steady state control of a serial line and transient control of a single pipe. Genetic algorithms are improvement algorithms modeled after the mechanics of natural genetics. They combine a survival-of-the-fittest mechanism with a structured, yet randomized, information exchange to search complex spaces quickly. In both pipeline problems, the genetic algorithm finds near-optimal results quickly without special programming, derivatives, or the restrictions of other methods.; Second, a learning classifier system (LCS) is applied to control two systems: an inertial object and a gas pipeline subjected to normal and leak operations. An LCS is a system that learns string rules for high performance behavior in arbitrary environments; it combines rules called classifiers, a universal message list, an apportionment of credit algorithm modeled after a competitive service economy, and a genetic algorithm. In the inertial object domain, restoration and braking rules are learned to center and stop the object after disturbance from rest. In the pipeline domain, the system learns rules for both normal operations and leak detection.; The genetic optimization and learning classifier system work are ready for near-term applications in pipeline operations. Genetic algorithms are ready for optimization in arbitrarily configured systems. Learning classifier systems, while still undergoing development, are ready for pilot applications in simple main line systems or stub lines. Ultimately, this work can lead to a computer system that acts as a consultant and a storehouse of pipelining knowledge. |
