Canal structure automation rules using a hydraulic simulation model, an accuracy -based learning classifier system, and genetic algorithms

• Main Author: Hernandez, Jairo E
• Format: Dissertation
• Language: English
• Published: ProQuest Dissertations & Theses 01-01-2008
• Subjects: Agricultural engineering; Artificial intelligence; Civil engineering
• ISBN: 9780549531098; 0549531092

Using state-of-the-art computational techniques, a genetic algorithm (GA) and an accuracy-based learning classifier system (XCS) were shown to produce optimal operational solutions for open-channel irrigation conveyance and distribution networks, based on objective functions supported by numerical flow analysis. An XCS successfully developed a set of operational rules for canal gates in an irrigation conveyance and distribution system through the exploration and exploitation of rules using a GA, with the support of an unsteady hydraulic simulation model. A computer program written in the C# .NET language solved the problem of how to operate all gate structures in an irrigation system simultaneously, while maintaining all water levels near their target values during variable demand periods, and afterwards with high hydraulic stability. Program inputs include the conveyance and distribution network layout and dimensions, adjustable control structures, and water demand hydrographs. Software runs, involving thousands of simulations each, were used in the testing and production phases of this research. More than one thousand hydraulic simulations were required to solve one problem, producing operational rules that maintained water levels along the network inside a dead band of ±8% of the target level depth. For the tested cases, from one hundred to two thousand XCS runs, each involving multiple hydraulic simulations, were required to produce satisfactory rules. The model converged for strong, feasible gate operational rules 100% of the time in the tested cases. The dimensionless rule fitness increased monotonically as runs progressed. Initial fitness started at an arbitrary value of 0.001, and satisfactory rules (which helped to achieve operational objectives) increased in strength during the training process. Fewer iterations were required to increase the fitness by increments of 0.001 as the classifier population evolved. A new parameter was defined as the cumulative absolute deviation from target levels per run, and it decreased as the model approached the final solution. Stability indexes approached zero not only during the process, but after all demands were satisfied and became constant. Water levels approached the respective target levels, and they became closer to the respective targets when the delivery demands became constant after an initial period of variable demand. In principle, the model can be applied to network layouts with constant or variable demands and is for universal use, within the limits of current hydraulic simulation capabilities. The output is a set of recommended operating rules for multiple flow control structures, facilitating the automation of open-channel irrigation conveyance and delivery systems. As a result, feasible and optimal rules are obtained, as was demonstrated by simulated system performance. Canal operators, designers, and irrigation system managers can benefit indirectly from this modeling tool, which is currently usable only by researchers, by enabling operational improvements for available water resources.