貨櫃吞吐量對台灣經濟扮演著舉足輕重的角色,且能帶動國家整體的經濟發展,因此建構未來貨櫃吞吐量需求模型,以提供足夠的資訊,做為政策研擬及港埠發展規劃之用,為當前重要的課題。本研究以單變量ARIMA模式、類神經網路及灰色GM(1,1)模型三種預測模型應用於高雄港貨櫃吞吐量之預測,且透過MAPE、RMSPE與THEIL三種評鑑指標加以評估其效能。研究結果顯示各預測模型均有高度之精準度,其中又以灰色GM(1,1)模型有最佳的適配能力,其次為單變量ARIMA模式,最後為類神經網路。
The container throughput plays an important role in Taiwanese economic and it is able to enhance national economic development. To provide the sufficient information for policy making and development planning of ports, building the future demand model of container throughput is a very important issue. This study uses univariate ARIMA, neural network, and GM (1,1) models to predict the container throughput in Kaohsiung port and uses MAPE, RMSPE and THEIL evaluation index to evaluate their efficacy. The analytical results show the predicting models are with high accuracy. The GM(1,1) model is the best for predicting container throughput, followed by univariate ARIMA and neural network.