短期列車旅運需求之類神經網路預測模式建構與評估

Artificial Neural Networks for Short-Term Railway Passenger Demand Forecasting

蔡宗憲(Tzung-Hsien Tsai) ； 李治綱(Chi-Kang Lee) ； 魏健宏(Chien-Hung Wei)

《運輸計劃季刊》 35卷4期 (2006/11) Pp. 475-505

https://doi.org/10.6402/TPJ.200611.0475

摘要

短期列車旅運需求預測模式可以提供軌道運輸營運者短期旅運資訊，有益於短期營運規劃之設計。本研究以類神經網路為基礎，建構短期列車旅運需求預測模式，探討三個模式構建課題：模式輸入變數設計對預測績效之影響、類神經網路模式與其他方法之績效比較、綜合模式(combined model)對預測績效之影響。我們蒐集臺鐵實際售票紀錄來進行模式構建以及驗證，有以下發現：其一，不適當的變數選用會導致類神經網路預測績效惡化。其二，在預測績效表現上，類神經網路優於隨機模式、去季節化隨機模式以及移動平均模式，但與指數平滑法相近。其三，綜合模式績效優於個別預測模式惟必須挑選較佳的個別模式來建立綜合模式。

關鍵字

短期旅運量預測 ； 類神經網路 ； 預測模式比較 ； 綜合模式

並列摘要

Short-term railway passenger demand forecasting can offer essential information to benefit short-term operational planning. This study constructed short-term forecasting models for railway passenger demand and discusses three modeling issues: the effects of input design on forecasting performance, validity of artificial neural networks and validity of combined models. We collected data from Taiwan Railway Administration for model construction and validation. Three findings were obtained. First, inappropriate design or use of input variables may result in unsatisfactory forecasting performance. Second, Artificial Neural Networks outperform random walk model, deseasonalized random walk model and moving average model, but have similar performance to exponential smoothing model. Third, combined models outperform individual models. However candidates should be carefully selected for combining.

並列關鍵字

Short-term forecasting ； Artificial neural networks ； Model comparison ； Combined model

參考文獻

魏健宏｜林士傑｜李穎(2003)。高速公路客運車輛旅行時間預測之實證評析。運輸計劃季刊。32(4)，651-680。

Adya, M.,Collopy, F.(1998).How Effective Are Neural Networks at Forecasting and Prediction? A Review and Evaluation.Journal of Forecasting.17,481-495.

Armstrong, J. S.(2001).Principles of Forecasting.Boston, US:Kluwer Academic Publishers.

Bazaraa, M. S.,Sherali, H. D.,Shetty, C. M.(1993).Nonlinear Programming: Theory and Algorithms.New York:Wiley.

Chatfield, C.(2000).Time Series Forecasting.USA:Chapman & Hall/CRC.