Measuring, monitoring and forecasting system for carbon monoxide concentrations

Loading...
Thumbnail Image
Date
2005
Authors
Raji, Usharani
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The impact of air pollution is broad towards the health of human beings, thus many studies have been focused on the forecasting of air pollutants in urban areas. In this research, a portable carbon monoxide forecaster has been developed in order to forecast the future level of carbon monoxide (CO) concentrations, since CO is primary pollutant which affects the air quality level. First, a PC based CO forecaster is developed in order to choose the best neural network and forecasting model. The chosen network together with its modeling will be later implemented in the portable system. From the comparison study, Hybrid Multilayered Perceptron (HMLP) network is chosen since it is found to perform better compared to Multilayered Perceptron (MLP), Recurrent and Radial Basis Function (RBF) networks, respectively. In this study, the HMLP network is trained with Modified Recursive Prediction Error (MRPE) algorithm to perform CO concentrations level forecasting. Besides that, only past CO concentration values are used as network input series, in order to perform forecasting. Meanwhile, on-line model is chosen since it is found to perform better compared to off-line model. This is due to its flexibility to update the network parameters for each data sample. The portable CO forecaster is developed by using dsPICDEM 1.1 development board, which uses digital signal controller dsPIC30F6014 device. The portable CO forecaster is able to perform CO concentrations measurement and forecasting up to 8 steps ahead. Besides that, the portable CO forecaster has the ability to store the measurement and forecasting values, in order to be transferred to PC through serial communication later on. HMLP network together with on-line model is implemented into the architecture of portable system. The developed portable CO forecaster is evaluated by using 3 real data sets and 2 simulated environment data sets, respectively. The R2 values achieved by these data sets for OSA test and 8 steps ahead forecasting are 0.9805 and 0.8501, 0.9677 and 0.7910, 0.8777 and 0.1524, 0.9966 and 0.8994, 0.9942 and 0.4097, respectively. The portable CO forecaster only uses 5 past CO concentration values, thus it can be concluded that the portable system gave excellent results over those data sets. Overall, it can be concluded that the developed portable CO forecaster can provide accurate and high accuracy of CO concentrations forecasting.
Description
Master
Keywords
Industrial technology , Forecasting System , Carbon Monoxide
Citation