Exploration of the search space of the in-core fuel management problem by knowledge-based techniques
暂无分享,去创建一个
The process ongenerating reload configuration patterns is presented as a search procedure. The search space on the problem is found to contain ∼10 12 possible problem states. Incomputational resources and execution time necessary to evaluate a single solution are taken into account, this problem may be described as a «large space search problem.» Understanding on the structure on the search space, i.e., distribution on the optimal (or nearly optimal) solutions, is necessary to choose an appropriate search method and to utilize adequately domain heuristic knowledge. A worth function is developed based on two performance parameters: cycle length and power peaking factor. A series on numerical experiments was carried out; 300 000 patterns were generated in 40 sessions. All these patterns were analyzed by simulating the power production cycle and by evaluating the two performance parameters. The worth function was calculated and plotted. Analysis on the worth function reveals quite a complicated search space structure. The fine structure shows an extremely large number on local peaks: about one peak per hundred configurations. The direct implication on this discovery is that within a search space on 10 12 states, there are ∼10% local optima. Further consideration on the worth function shape shows that the distribution on the local optima norms a contour with much slower variations, where «better» or «worse » groups on patterns are spaced within a few thousand or tens of thousands of configurations, and finally very broad subregions of the whole space display variations of the worth function, where optimal regions include tens of thousands of patterns and are separated by hundreds on thousands and millions. The main conclusion is that the basic challenge on the reload configuration design is due to an extremely large search space and its complicated structure. Heuristically guided search seems to be well suited nor this problem
[1] A. Galperin,et al. Application of knowledge-based methods to in-core fuel management , 1991 .
[2] Geoffrey T. Parks,et al. An Intelligent Stochastic Optimization Routine for Nuclear Fuel Cycle Design , 1990 .
[3] Alexander Sesonske,et al. PRESSURIZED WATER REACTOR OPTIMAL FUEL MANAGEMENT. , 1972 .