Genetic Programming Approach to Data Mining and Knowledge Discovery from Store Separation Trajectories

A data mining and knowledge discovery (DMKD) methodology based on Genetic Programming is proposed for stores separation trajectory analysis. The proposed technique constructs symbolic relationships between a defined set of input variables and analysis variables using the trajectory data. These relationships can be used by store separation trajectory analysts to gain a deeper understanding of the information contained in the trajectory database. They can be used to quickly survey trajectories obtained from sources such as Monte Carlo simulations, genetic algorithm searches and flight tests, and provide important trends and direct attention to areas requiring attention. These symbolic relationships will help assist store separation analysts to select inputs for Monte-Carlo simulations to ensure that unacceptable trajectories are avoided, and will help interpret the behavior of extreme trajectories. Phase I research will develop a preliminary prototype of the DMKD software designed to work in a well-known numerical environment. Trajectory data supplied by the Air Force will be used to demonstrate the capabilities of the data mining software. Phase II research will develop the full version of the DMKD software. The DMKD techniques and the trajectory analysis software developed under the present SBIR project will be commercialized during the Phase III work.