An Application of Chaos Dynamics to the Analysis and Prediction of Surface Ozone Concentrations in Urban Area

• Main Authors: Yu-Chi Weng, 翁御棋
• Other Authors: Ni-Bin Chang
• Format: Others
• Language: zh-TW
• Published: 2001
• Online Access: http://ndltd.ncl.edu.tw/handle/02868081863251421201

碩士 === 國立成功大學 === 環境工程學系 === 89 === The proper forecasting of ground-level ozone concentration frequently plays an important role in environmental quality management in the metropolitan region. Though ozone model prediction models exist, there is still a need for more accurate models. Many different forecasting models have been applied for the prediction of ground-level ozone concentration in the last decades. But none of them is capable of exploring the chaos characteristic embedded in the atmospheric environment. Since the chaos property has been regarded as a dominant factor in the meteorological phenomenon for at least over ten years, it is the aim of this study to explore the possible linkage between chaos property and ozone prediction. This analysis starts from introducing the theory of fractal and chaos, proceeding with a search for the well-known indexes in chaotic dynamics, such as the Lyapunov exponent, the Hurst exponent, etc., and ending up with a case study via collecting the monitoring database of Zen-Wu and Chow-Chu in the Kaohsiung metropolitan region for further assessment. Model comparison, including time series ARIMA model, neural network model, and non-linear chaotic forecasting model, was also conducted simultaneously for final appraisal. The BDS test and Rescale Range (R/S) analysis provided evidence for non-linearity and fractality due to noisy chaos, respectively, in this analysis. It is verified by the BDS test that ground-level peak ozone concentration exhibits strong non-linearity characteristic. The Hurst exponent estimated with respect to the monitoring data of ground-level peak ozone concentration at Zen-Wu and Chow-Chu is 0.8441 and 0.8286, respectively. According to the V statistics derived for these two monitoring stations, it appears that non-periodic cycle lengths are apparent in 32 day, 170 day, and 420 day simultaneously. Methods of chaotic dynamics, like correlation dimension and related tests, and Lyapunov exponents, give consistent results, which do not rule out the possibility of deterministic chaos. The low-dimension attractor derived at Zen-Wu and Chow-Chu is 5 in the sense that at least 5 dynamic variables are required to depict such dynamic system. Nevertheless, which 5 variables are required has not yet clear. In any circumstances, the chaotic property is apparent in the data analysis of ground-level peak ozone concentration in this area. The integration of chaotic prediction with the aid of neural network model not only exhibits accurate short-term forecasting capability but also illustrates the non-linear, chaotic property in applications.