Using Two-Stage Data Mining in Prediction of Demand and Return for the YouBike Station

• Main Authors: LIN, KUAN-TING, 林冠廷
• Other Authors: HOU, TUNG-HSU
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/r9gve5

碩士 === 國立雲林科技大學 === 工業工程與管理系 === 106 === In order to achieve sustainable development and implement the concept of energy saving and carbon reduction, public bicycles have gradually replaced some of the means of transportation since 2009. The YouBike developed by Giant has main six cities in Taiwan, but some county also has their own public bicycle systems. At present, the cities and counties to supplement the bicycle, often according to the previous experience, there is no a general system of complement. Therefore, this study will use data mining technology to 1) establish a prediction model; 2) explore the next period of the site demand and the amount of return forecast. The data used in this study base on YouBike data using history, between 2016 to 2017 Changhua train station rental information for analysis. Because of the time series of demand and return, the traditional back-propagation neural network cannot be used to predict. First stage data mining used adaptive resonance theory neural network identified each site requirement type and input second stage data mining. Then time-delay neural network and recurrent neural network are used for training, testing, and using autoregressive integrated moving average models. Three different models are compared to find a suitable prediction method. The results show that the use of clustering algorithm to effectively find out the similar demand / return pattern, and then with the time-delay neural network prediction can get better forecast performance. The results of the study can be combined with the real-time monitoring system to make decisions in the future.