中高速船の横切面積曲線計画ツールの構築

ニューラルネットワークを利用した船型初期計画 (第2報)

抄録

In this paper, the nonlinear programming system is proposed for preliminary planning of area curve for high speed ships. The system is composed of 4 neural networks ; the Estimating Net, Descriptive Net, Integrating Net, and Optimizing Net.
The Optimizing Net proposes an area curve based on the design condition, i. e. target Froude number F_n, target Prismatic coefficient C_P, Length Breadth ratio L/B, and Breadth draft ratio B / d. Its output is connected to the inputs of other three forward model networks, which calculate resulted C_P, wave making resistance coefficient r _w, and teaching-data density parameter δ. The Optimizing Net is adjusted to minimize the evaluation function J, whose elements are the outputs of three forward model networks using the error-back propagation method. Finally, the Optimizing Net proposes the optimum area curve for the given design condition.
Since the definition of J includes the output of Descriptive Net, the resulted area curve is certificated that it is inside of the domain of teaching data which are adopted for generation of the forward model networks. The introduction of Descriptive net's output to the evaluation function is one of the important features of the proposed system.
The proposed optimizing system involves non-linear effects on design parameters and their crossinteraction, so that it could recommend an innovative idea for ship design.
It should be emphasized that the forward model networks were generated based on the tank test data of Series60. When the other data are available, it is possible to make the system to cover much wider variety of ship hull form.