Premium
Application of a neural network to a design optimization process
Author(s) -
Zandinia A.,
Koppelaar H.
Publication year - 1992
Publication title -
international journal of intelligent systems
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.291
H-Index - 87
eISSN - 1098-111X
pISSN - 0884-8173
DOI - 10.1002/int.4550070804
Subject(s) - computer science , heuristics , travelling salesman problem , artificial neural network , norm (philosophy) , artificial intelligence , process (computing) , plan (archaeology) , mathematical optimization , network planning and design , theoretical computer science , operations research , mathematics , algorithm , computer network , archaeology , political science , law , history , operating system
This article discusses the implementation of a class of design problems in a neural network. the problems, identified at an abstract (i.e., connectivity) level of building design, are characterized as optimization types of problems. Architects often have to generate floor‐plan layouts of a building optimized with respect to several points of view. Examples of these points of view, often stated in terms of normative requirements, are the social norms: community, privacy , and circulation‐cost. an architectural design problem in the presence of even a single norm from among the above norms is computationally hard and intractable. During the last four decades there have been attempts to automate floor‐plan design considering a single norm or a limited number norms. Most of these attempts have traditionally been based on combinatorial enumeration methods. Recent progress in AI has paved the way for intelligent handling of the architectural design processes using knowledge‐based system technology and heuristics programming. This article examines the possibility of the neural networks approach in generating connectivity patterns of building with respect to specific social norms. Our first attempt has been focused on linear‐tree type designs with respect to single norms. the idea in this work was borrowed from the Hopfield model of the neural network for implementation of the Traveling Salesman Problem, because of the similarity of our design problems with this problem. Hopfield and other researchers used constant parameters for different‐sized problems. Close examination of the network and experiments revealed that this approach does not guarantee a convergence for every case, and chaotic behavior is expected in cases for which the chosen parameters are not appropriate. to overcome this problem this article suggests problem‐dependent and problem‐size‐dependent parameters that vary for each case. Test results from the implementation convey that the approach yields satisfactory results and is worth it to explore its application to other classes of optimization problems.