Professor, CSE
Director, ECSL
Book Chapters
This paper investigates case-initialized genetic algorithms for extracting knowledge from past problem solving to solve subsequent problems. We develop a test problem class with similar solutions and the genetic algorithm is run for randomly chosen problems from the class. As the algorithm runs on a particular problem, solution strings are stored in a case-base and on subsequent problems, solutions from the case-base are used to initialize the population of a genetic algorithm. We investigate the effect of selection strategy and choice of appropriate cases for injection. Scaled roulette and scaled elitist selection both show improvement over a randomly initialized GA and elitist selection performs better than roulette. Over 50 problems the case-initialized genetic algorithm system shows a statistically significant decrease in the time taken to the best solution and solutions are of higher fitness. Several strategies for choosing cases from the case base for injection all provide measurable improvement over random initialization.
We report on a spectroscopic method for the characterization of the spatial structure of inertial confinement fusion implosion cores based on the self-consistent analysis of simultaneous narrow-band X-ray images and X-ray line spectra. The method performs a search in multi-dimensional parameter space for the temperature and density gradients that simultaneously yield the best fits to narrow-band spatial emissivity profiles obtained from X-ray images, and spectral line shapes recorded with crystal spectrometers. A multi-objective Niched Pareto Genetic Algorithm (NPGA) was developed to efficiently implement the multi-criteria data analysis. The availability of the NPGA is critical for the practical implementation of this analysis method, since NPGA-driven searches in parameter space typically find suitable solutions in approximately 105 evaluations of the spectral model out of a total of 1018 possible cases (i.e. size of the parameter space). Furthermore, analysis of solutions on the Pareto front permits us to address the issue of uniqueness of the solution and the uncertainty of the optimal solution. The performance of the NPGA is illustrated with spectroscopic data recorded in a series of stable and spherically symmetric implosion experiments where argon doped deuterium-filled plastic shells were driven with the GEKKO XII (Institute of Laser Engineering, Japan) and OMEGA (Laboratory for Laser Energetics, USA) laser systems. This measurement is relevant for understanding the spectral formation and plasma dynamics associated with the implosion process. In addition, since the results are independent of hydrodynamic simulations they are important for the verification and benchmarking of detailed hydrodynamic simulations of high-energy-density plasmas.