Summary: | 碩士 === 國立成功大學 === 環境工程學系 === 88 === Meteorological forecasting is an essential tool to present accurate prediction of air pollution concentration from a regional sense. It is the aim of this study to test the possibility and the potential via the use of neural networks model as a means for generating the diagnostic wind field as the conventional mass-consist flow field is performed as a comparable basis for advanced verification. With the support of particle model and puff model, the real-time prediction of air pollution concentration may become achievable. The merit of such applications is placed upon the real-time forecasting of the potential impact of accidental release from nuclear power plants due to insufficient weather data that can be collected and transfer within very short period of time. Key issues related to such development include how many number of meteorological, topological, geographical, and environmental factors can be covered in the neural networks model and how to build up such a model with an optimal structure via a numerical procedure. The main findings of this study include:
Neural networks model is a feasible model to be used as a means with a reasonable speed of convergence in the numerical procedure.
The use of neural networks model may generate relatively reliable wind field pattern although statistics show that conventional mass-consist flow field presents relatively higher accuracy in terms of both the wind speed and wind direction.
The performance of particle model is generally better than that of puff model due to its higher sensitivity to topological variations.
Based on three different representative weather patterns in North Taiwan, case study, using the first nuclear power plant as an example, clearly indicates that accidental release in the most severe condition could only generate direct impacts to local village but can only create very limited impacts to the neighboring highly populated region, such as Pai-Tou and Dan-Sui.
|