Neural network modeling of ozone and particulate matter concentrations in Southeast Texas.

摘要

In this thesis study the neural network modeling technique is used to model the daily ozone concentration buildup and the hourly particulate matter concentrations in Southeast Texas. The ozone model was to predict the increase of daily ozone concentration increase based on the 9 am conditions. The model has five input parameters: temperature, solar radiation, nitric oxide, nitrogen dioxide and wind speed. The neural network analysis was done for Beaumont (C02), Hamshire (C64), Sabine Pass (C640) and Port Arthur (C643) sites in the BPA region. The particulate matter model was to predict hourly PM concentration based on various sets of selected input parameters. These parameters included hourly data of nitric oxide, nitrogen dioxide, sulfur dioxide, oxides of nitrogen, wind speed, temperature, local wind direction, regional wind direction, hour (time of the day) and ozone. The neural network analysis was done for the Beaumont (C54) site. The neural network analysis was performed using the Brain Maker Professional from California Scientific Software. The results for the ozone model indicate the R-square values are high and that the effects of the various parameters are specific to the individual sites. The results for the particulate matter model indicate that the R-square values are relatively low implying the model needs further improvement.