Forecasting Model for Automobile in Taiwan-Performance Comparison between ARMAX and Artificail Neural Network

• Main Authors: Chih-YuanCheng, 鄭智遠
• Other Authors: Kuo-Ping Huang
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/27652740166793690608

碩士 === 國立成功大學 === 交通管理學系碩博士班 === 101 === Recently, the issue of environmental sustainability is more important than before, most of countries are really focusing in this issue. Depleted resource and global warming are becoming the important issue for human living. The volume of CO2 release is one of the important part for sustainability of development. People cannot slight energy deplete and CO2 release from cars, therefore people have to master the growth of vehicle number and its factors. This thesis use two different patterns to create vehicle ownership’s models, ARMAX and ANNs. Base on compare performance for this two models, we can predict the trend of number of car growth and the relationship between different variables to understand what factors will affect the number of vehicle ownership. In this thesis collected number of vehicle and motorcycle ownerships in Taiwan area from 1981 to 2010 and real GDP, exchange rate(EXCH), price of Brent Blend Crude Oil, population, employment rate and number of house’s yearly data. Also using dummy variables to present in 1987 Taiwan implemented licenses renewal, in 1999 Taiwan implemented investigation of vehicles and motorcycles, in 2008 economic crisis and in 1995 license tax free for motorcycle under 150CC. The estimated results of ARX and ARMAX vehicle model show that vehicle ownership can be explained by the real GDP, Transportation and communication CPI(TCCPI), and Taiwan implement investigation for vehicles and motorcycles in 1999(D2). The estimated results of ARMAX motorcycle model show that vehicle ownership can be explained by the real GDP, Transportation and communication CPI(TCCPI), and license tax free for motorcycle under 150CC in 1995 (D4). From sensitive test for ANNs, income is most sensitive in all factors, then is population factor, the last one is price factor. In the accuracy of prediction, compare models of ARMAX(MAPE is 3.9040) and ANNs(MAPE is 3.9631), ARMAX have higher skill of prediction value and it also have skill of explanation.