Toxic Gas Dispersion Model Based On Neural Pattern Recognition Networks
Loading...
Date
2022-07-01
Authors
Roslan, Nurfarah Arina
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Sains Malaysia
Abstract
The chemical engineering industry has grown steadily for the past few years that causes many catastrophic incidents involving chemical industries. Prairie Grass experiment database is used as a data to develop toxic gas dispersion prediction model based on deep learning networks. Thus, in this study, development of deep neural network is carried out using MATLAB. There are 14 parameters consist of 6583 samples related to toxic gas dispersion from Prairie Grass experiment is used. To achieve the objectives, two phases of structure architecture of NPR is carried out. First, NPR development is developed using three different algorithms which are Levenberg-Marquart (LM), Bayesian Regularization (BR) and Scaled Conjugated Gradient (SCG) to propose the best network algorithm using 70% training and 10-28 hidden neurons. From the analysis, BR shows the best network algorithm compared to others by giving maximum R-value of 0.95. Following the best selection of neural network algorithm, BR algorithm is further trained using 50-70% training with 10-28 hidden neurons. As a result, BR algorithm using 70% training and 28 hidden neurons give the best performance with R-value of 0.95214. Thus, the NPR model is a reliable model for toxic gas dispersion model.