In designing the nuclear reactors there are some independent variables the values of which can be selected by the designer. The power generation cost, for example, depends upon the values of these design parameters and, hence, optimization has to be carried out to find out the value of the parameters for which the cost is minimum. This type of problem has been solved by the designers relying on their engineering judgement. Recent developments in systems engineering would appear to afford a more sophisticated approach to the problem.
In this paper, optimization methods are applied to a problem of this kind; the heavy water moderated, boiling light water cooled reactor has been selected as an example. The cost is too complicated to be described in a closed form and, therefore, analytical methods of optimization are difficult to use. Two typical direct search methods, namely the pattern search method and the simplex method, are applied. Both methods have proved to give reasonably good results. But the cost factor in this problem is multi-modal, and hence sometimes only a local optimum is obtained instead of the global, and a point very close to the optimum is easily discovered, but much computation is required to ascertain that this point is close to, at least a local, optimum.
From the above results it is concluded that these methods are useful for optimizing the reactor design in the stage of "parametric survey", and that if we can develop effective ways to find the global optimum and to ascertain that a point is close to the optimum, then these methods can become even more useful.
抄録全体を表示