Page 380 - Practical Design Ships and Floating Structures

355

(4 (b) (c) (4
Figure 6: Predictions of displacement and draught of a large container vessel in terms of
probability distributions. (a) and (b) are restricted in TEU capacity, whereas (c) and (d)
are restricted with respect to TEU capacity and Panmax dimensions.

It is further specified that the ship must not exceed the maximum dimensions of the Panama Canal
(L 5 294.2 m, B 5 32.3 m, D 5 13.4 m). This information can be inserted in the Bayesian network in
the form of likelihood vectors. The dimension of the vector is equal to the number of states of the
corresponding variable, and its elements are zero for impossible (unwanted) states and unity for
possible states. When the new evidence is propagated the probability distributions are updated again.
This does not change the distribution much for the displacement (Figure 6(c)), but as the draught is
restricted its distribution changes, Figure 6(d). The Panmax requirements are not fulfilled by the
estimates produced by the single input neural network, so a network with fixed TEU and breadth as
inputs is trained. The results for displacement and draught are shown in Figures 6(c) and 6(d),
respectively. With both TEU capacity and breadth specified, the neural network produces smaller
variations in the predicted displacement and draught than the single input neural network.
4.2 Smalc Container Vessel

As another example, a smaller container vessel is considered. On the assumption of a capacity of
approximately 950 TEU, evidence is inserted in the Bayesian network in the interval [846, 10621 and
after propagation, the distributions of L and D are as shown in the bar charts of Figures 7(a) and 7(b).
Here, a length between 140 and 160 m and a draught between 8 and 8.7 m are estimated to be the most
probable. The predictions of a single-input neural network and the simple regression for the same
range of TEU capacity are also shown.

By further requiring that the draught is limited to 8.5 m, the distributions from the Bayesian network of
length and draught are as shown in the bar charts in Figures 7(c) and 7(d). The length and draught
intervals mentioned above are now estimated to be even more likely. A neural network with capacity
and draught as input is trained. The draught is set to 8.5 m and the result of inserting 846 TEU and
1063 TEU in the neural network is shown in Figures 7(c) and 7(d) as 'neural network, double input'. It
should be noted that the evidence inserted in the Bayesian network (D < 8.5 m) is not fully equivalent
to the input into the neural network (D = 8.5 m). This illustrates the flexibility of Bayesian networks.

I

Rcnerrn
Ez.Wz%
*
I 4 8012",It*
m.8. -1 L-UII11 aM *I
(a) (b) (4 (4
Figure 7: Predictions of length and draught of small container vessel. (a) and (b) are restricted
in TEU capacity, whereas (c) and (d) are restricted with respect to TEU capacity and draught.