Computational & Technology Resources
an online resource for computational,
engineering & technology publications |
|
Civil-Comp Proceedings
ISSN 1759-3433 CCP: 82
PROCEEDINGS OF THE EIGHTH INTERNATIONAL CONFERENCE ON THE APPLICATION OF ARTIFICIAL INTELLIGENCE TO CIVIL, STRUCTURAL AND ENVIRONMENTAL ENGINEERING Edited by: B.H.V. Topping
Paper 20
Simulating Ozone Level Time Series using an Innovative Hybrid Model based on a Multilayer Perceptron D. Wang and W.Z. Lu
Department of Building and Construction, City University of Hong Kong, Kowloon, Hong Kong D. Wang, W.Z. Lu, "Simulating Ozone Level Time Series using an Innovative Hybrid Model based on a Multilayer Perceptron", in B.H.V. Topping, (Editor), "Proceedings of the Eighth International Conference on the Application of Artificial Intelligence to Civil, Structural and Environmental Engineering", Civil-Comp Press, Stirlingshire, UK, Paper 20, 2005. doi:10.4203/ccp.82.20
Keywords: tropospheric ozone level time series, ozone episodes, the hybrid model, hybrid monte carlo, multilayer perceptron, particle swarm optimization.
Summary
To produce reliable and accurate forecasting for tropospheric ozone (O3) levels is
a challenge and significant task related to public health [1]. In this study a multilayer perceptron
(MLP) model is trained with the original particle swarm optimization (PSO) algorithm,
and named as the MLP-PSO model. It can be used to perform such a task [2,3],
but the PSO algorithm tends to suffer from the "curse of dimensionality" problem, when the
architecture of the MLP becomes large [4]. Such a difficulty inevitably degrades the model's
predictive performance. A proper method for the initialization of the swarm of the PSO can
alleviate such a difficulty [5,6,7].
In this study, the hybrid Monte Carlo (HMC) method is employed to sample the weight matrix from the posterior probability distribution of the MLP optimal weight matrix [8,9,10], and the sampled weight matrix is then used to initialize the "weight matrix swarm" in the PSO before the MLP training starts. The MLP-PSO with this new swarm initializing strategy can be considered as a two-staged hybrid model and the selecting of the model parameters are based on the general recommendations given in the literature for the HMC and the PSO, as well as determined by the try and error method [11,12,13,14,15]. Before the predictive experiments, an experiment aiming to compare the initial swarm (particles) (one is sampled by the HMC and the other is randomly distributed) is conducted. The result implies that the swarm sampled by the HMC method has a better convergence rate and reliability for the optimal weight matrix than the one obtained by the random scheme. Within expectation, the visualized experimental results of a 1-day ahead forecast for the daily maximum O3 level in two selected air monitoring sites, shows that the hybrid model can provide more accurate predictions, especially in episodes [16,17]. Further comparison for the root mean square error (RMSE) of the two models for each season and each site reveals that the level of performance degradation for the MLP-PSO model is obviously larger than that of the hybrid model in episodes. This indicates that the hybrid model is robust to give reasonable predictions for the complex variation patterns of O3. References
purchase the full-text of this paper (price £20)
go to the previous paper |
|