| Summary: | 碩士 === 國立成功大學 === 交通管理學系碩博士班 === 90 === As fast progresses in science and technology development and income growth have been achieved in Taiwan in past decades, traffic situation became worse and worse. The motorcycle here in Taiwan is also extremely large in number. This phenomenon causes local traffic problems more complicated. To achieve the goal of sound motorcycle management, we need to understand the characteristics of local motorcycle traffic flow.
As the motorcycle is not a main transportation mode in overseas countries, the development of traffic flow models in abroad has been confined only to automobiles. Very few studies were focused in motorcycle traffic. Based on this understanding, this study tries to collect data via photographing the real-world motorcycle traffic flow from overhead in order to develop a motorcycle traffic flow model. By using Neural Network method, local characteristics and requirements are analyzed to build said model to become a base for the following motorcycle studies.
After detailed data analyzing, the motorcycle progress path on an excluded motorcycle way has been divided into two dimensions, i.e. longitudinal moving and traversal moving for its traffic model development. The longitudinal progress model developed from Neural Networks is found to have very good performance. On the other hand, the traversal progress model, owing to its random overtaking behavior from either side of its preceding motorcycles, does not perform well to capture the actual traversal path of target motorcycles.
To be the first one to employ Neural Network method for building the motorcycle traffic flow model, this study has completed the whole model-building process with parameter calibration, model verification and model validation. The main achievement of it is to confirm the feasibility of constructing the motorcycle traffic flow model by using Neural Networks as well as its practical usage in the real-world motorcycle traffic flow conditions.
|
|---|
