Page 338 - Practical Design Ships and Floating Structures

3 13

compartment layout design case. So the network showed in Fig.2 has characteristics as following:

1. Input layer composes of two sets of unit Il and I,. The two sets of unit have the same unit number
and same unit sequence. They express different corresponding properties of the compartment layout
design task.

2. Output layer have only one unit. The value of the unit is the degree of similitude of two
compartment layout design cases inputted. The assign of weight is done by neural network. Supposing
we have n pieces of compartment layout design cases and each case have m pieces of property which
can be decomposed in the universal set of layout design task, the neural network would have 2m pieces
of input unit. The design task properties of every two compartments layout design cases and the degree
of similitude of these two compartment layout design cases construct a sample. The first one is the
input, and the second is output. If we call the collection of design task property value a as C,
CI=[a,,, q2, -, a,,], i=l, 2, , n. If the input of the sample is I, I, = [a,, .a*, a,,,
n.
a,, , uj2 ;-, a,m 1, i, j = 1,2, --, If0 is the output of sample, Eqn. (2) is the corresponding matrix
of sample input and output. In these matrixes, elements in corresponding position constitute a pair of
samples.

Iz2, I, '.. 12, 1 # 102, 022 ." 02,
. . . . . . . . . . . . . . . . . . . . . , . .

-
1 01, 013 ... 01,
1 0, .'. o,,
1 ... ..

1 O(n-l)n
- 1