Applying Constraint Satisfaction Neural Networks to Multiple Views' Data Partitioning in Building Engineering

by D. H. Douglas Phan, Stanford Univ, Stanford, United States,

Document Type: Proceeding Paper

Part of: Computing in Civil and Building Engineering


A typical building engineering process involves multiple project participants and several computer applications. These participants and applications have their own views of the underlying building data Since these users (i.e., participants and applications) contribute to and draw from the same overlapping set of data. the problem here is to partition the set of building data into logical fragments that meet two major requirements: (1) supporting those multiple views and (2) still being meaningful to the domain. I call this problem the data partitioning problem. It is analogous to the vertical fragmentation problem in distributed database design, for which algorithms using matrix manipulation were proposed (Ozsu, 91). The limitations of these algorithms are that they address only the first requirement and are computationally expensive. In this paper, I propose constraint satisfaction neural networks that handle both requirements in solving the problem. The user, in this case, is the designer of the database supporting those participants and computer applications. As a conclusion, a database designer can use these neural networks as effective assistant tools to partition building data in support of multiple views or data uses. In each problem occurrence. these networks yield a high probability of finding the best solution. Moreover, the designer has direct control over the way in which each network goes about solving the problem. This capability comes from a number of parameters whose value can be specified by the designer. By changing the ratio of their values, the designer can define a set of prioritized criteria and can watch the solution that the network comes up with each time. Each solution comes with a measure of how well it satisfies the problem constraints according to the given criteria. The designer can compare all these solutions based on their measure and can then make a conscientious decision about which solution to use.

Subject Headings: Neural networks | Computing in civil engineering | Buildings | Architectural engineering | Databases | Building design | Algorithms | Structural design

Services: Buy this book/Buy this article


Return to search