| 摘要 |
PROBLEM TO BE SOLVED: To easily obtain the rationalization of design data by means of a neural network. SOLUTION: An estimation network which estimates and outputs wave making resistance factors by using design parameters specifying the type of a ship as a input layer 10, an associated network 14 which outputs the density of learning points in a space formed of design parameter groups, and an integration network 16 which calculates pilaster factors from the parameters related to the shape of a cross-cutting area curve selected from the design parameters are built. In addition, an adjustment network 18 which uses minimum design condition parameters as the input layer and the design parameters as an output layer is built in the preceding stage of the input layer 10. The estimation, associated, and integration networks 12, 13, and 16 are made to learn by adjusting the coupling loads of the networks 12, 13, and 14 by inputting the actual result data to the design parameters and the error energy between design data and teaching data obtained based on the output of each network when the design data are inputted to the design parameter groups is calculated as a target function J. Then the coupling load of the adjustment network 18 is adjusted so that the error energy may become the smallest value.
|
