Scribd
Upload
69 views1 page

2.3.8 Parallel Neural Network Algorithms For Large-Scale Problems

Uploaded by

Abdul Rahman
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
69 views1 page

2.3.8 Parallel Neural Network Algorithms For Large-Scale Problems

Uploaded by

Abdul Rahman
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 1

12 OVERVIEW OF NEURAL NETWORKS IN CIVIL ENGINEERING

Ghaboussi et al. (1998) present autoprogressive training of neural network constitutive models using the
global load-deflection response measured in a structural test with application to laminated composites. In
their approach a partially trained neural network generates its own training cases through an iterative
nonlinear finite element analysis of the test specimen. Yeh (1999) uses the BP algorithm to model the
concrete workability in design of high-performance concrete mixture. Neural network is also used to model
generalized hardening plasticity (Theocaris and Panagiotopoulos, 1995), alkali-silica reaction of concrete
with admixtures (Li et al., 2000), and elastoplasticity (Daoheng et al., 2000).

2.3.8
Parallel Neural Network Algorithms for Large-Scale Problems
The convergence speed of neural network learning models is slow. For large networks several hours or even
days of computer time may be required using the conventional serial workstations. A parallel
backpropagation learning algorithm has been developed by Hung and Adeli (1993) and implemented on the
Cray YMP supercomputer. A parallel processing implementation of the BP algorithm on a Transputer
network with application to finite element mesh generation is also presented by Topping et al. (1997).
Optimization of large structures with thousands of members subjected to actual constraints of commonly-
used design codes requires an inordinate amount of computer processing time and high-performance
computing resources (Adeli and Kamal, 1993; Adeli, 1992a&b; Adeli and Soegiarso, 1999). Park and Adeli
(1997a) present a data parallel neural dynamics model for discrete optimization of large steel structures
implemented on a distributed memory multiprocessor, the massively parallel Connection Machine CM-5
system. The parallel algorithm has been applied to optimization of several highrise and superhighrise
building structures including a 144-story steel superhighrise building structure with 20096 members in
accordance with the AISC ASD and LRFD codes (AISC, 1995, 1998) and subjected to multiple loading
conditions including wind loading according to the Uniform Building Code (UBC, 1997). This is by far the
largest structural optimization problem subjected to actual constraints of a widely-used design code ever
solved and reported in the literature. Park and Adeli (1997b) present distributed neural dynamics algorithms
on the Cray T3D multiprocessor employing the work sharing programming paradigm.

2.4
ENVIRONMENTAL AND WATER RESOURCES ENGINEERING
Karunanithi et al. (1994) demonstrate the use of neural networks for river flow prediction using the cascade-
correlation algorithm. The BP algorithm is used by Du et al. (1994) to predict the level of solubilization of six
heavy metals from sewage sludge using the bio-leaching process, by Grubert (1995) to predict the flow
conditions at the interface of stratified estuaries and fjords, by Kao and Liao (1996) to facilitate the selection
of an appropriate facility combination for municipal solid waste incineration, by Tawfik et al. (1997) to
model stage-discharge relationships at stream gauging locations at the Nile river, by Deo et al. (1997) to
interpolate the ocean wave heights over short intervals (weekly mean wave heights) from the values
obtained by remote sensing techniques and satellites over long durations (a month), and by Liong et al.
(2000) for water level forecasting in Dhaka, Bangladesh.
Crespo and Mora (1995) describe neural network learning for river streamflow estimation, prediction of
carbon dioxide concentration from a gas furnace, and feedwater control system in a boiling water reactor.
Basheer and Najjar (1996) use neural networks to model fixed-bed adsorber dynamics. Rodriguez and
Serodes (1996) use the BP neural network to estimate the disinfectant dose adjustments required during

You might also like