Forecasting available parking space with largest Lyapunov exponents method
来源期刊:中南大学学报(英文版)2014年第4期
论文作者:JI Yan-jie(季彦婕) TANG Dou-nan(汤斗南) 郭卫红 BLYTHE T. Phil WANG Wei(王炜)
文章页码:1624 - 1632
Key words:available parking space; Lyapunov exponents; wavelet neural network; multi-step forecasting method
Abstract: The techniques to forecast available parking space (APS) are indispensable components for parking guidance systems (PGS). According to the data collected in Newcastle upon Tyne, England, the changing characteristics of APS were studied. Thereafter, aiming to build up a multi-step APS forecasting model that provides richer information than a conventional one-step model, the largest Lyapunov exponents (largest LEs) method was introduced into PGS. By experimental tests conducted using the same dataset, its prediction performance was compared with traditional wavelet neural network (WNN) method in both one-step and multi-step processes. Based on the results, a new multi-step forecasting model called WNN-LE method was proposed, where WNN, which enjoys a more accurate performance along with a better learning ability in short-term forecasting, was applied in the early forecast steps while the Lyapunov exponent prediction method in the latter steps precisely reflect the chaotic feature in latter forecast period. The MSE of APS forecasting for one hour time period can be reduced from 83.1 to 27.1 (in a parking building with 492 berths) by using largest LEs method instead of WNN and further reduced to 19.0 by conducted the new method.
JI Yan-jie(季彦婕)1, TANG Dou-nan(汤斗南)1, GUO Wei-hong(郭卫红)2, BLYTHE T. Phil2, WANG Wei(王炜)1
(1. Jiangsu Key Laboratory of Urban ITS, Southeast University, Nanjing 210096, China;
2. Transport Operations Research Group, School of Civil Engineering & Geosciences, Newcastle University,
Newcastle upon Tyne NE1 7RU, UK)
Abstract:The techniques to forecast available parking space (APS) are indispensable components for parking guidance systems (PGS). According to the data collected in Newcastle upon Tyne, England, the changing characteristics of APS were studied. Thereafter, aiming to build up a multi-step APS forecasting model that provides richer information than a conventional one-step model, the largest Lyapunov exponents (largest LEs) method was introduced into PGS. By experimental tests conducted using the same dataset, its prediction performance was compared with traditional wavelet neural network (WNN) method in both one-step and multi-step processes. Based on the results, a new multi-step forecasting model called WNN-LE method was proposed, where WNN, which enjoys a more accurate performance along with a better learning ability in short-term forecasting, was applied in the early forecast steps while the Lyapunov exponent prediction method in the latter steps precisely reflect the chaotic feature in latter forecast period. The MSE of APS forecasting for one hour time period can be reduced from 83.1 to 27.1 (in a parking building with 492 berths) by using largest LEs method instead of WNN and further reduced to 19.0 by conducted the new method.
Key words:available parking space; Lyapunov exponents; wavelet neural network; multi-step forecasting method