Abstract:

Road multi interval travel time forecasting data is an important parameter for dynamic traffic guidance system. But previously developed models have some deficiency, such as bad adaptability, large amount of calculation needing and many history data requirement. Applied the decomposition, reconstruction of spectral analysis and Karhunen-Loeve (K-L) transform method to decompose and reconstruct the history and current detection travel time series. Euclidean distance is used to measure the closeness between current and historical travel time series, by the means of optimization the eigenvector coefficient to make those more closely history series has the more weight in the reconstruction and then gain the goal of road travel time multi step forecasting. The case study suggest that, the proposed method can fulfill multi step road travel time prediction and has a good prediction accuracy, some better than the previous method, furthermore, a fewer history data and calculation resources needing.

Key words: urban traffic, travel time, multi-step prediction, Karhunen-Loeve transform, time series reconstruction

摘要：

路段多步行程时间预测数据是动态交通诱导系统的重要参数，但已有研究成果，大多集中于一步预测，且存在适应性不强、计算量大、基础数据需求多等不足.应用谱分析及Karhunen-Loeve（K-L）变换对历史及当前检测行程时间序列进行分解与重构，重构时以历史序列与当前检测序列的欧式距离作为相似性度量指标，优化重构时的特征向量系数，使与当前检测序列相似度高的历史序列信息在重构中占据主要地位，通过重构，实现对后续若干时段的行程时间的预测，实测数据检验显示该方法可实现多步预测，预测精度良好，较以往方法有所提高，且历史数据需求量小，计算量小.

关键词: 城市交通, 行程时间, 多步预测, Karhunen-Loeve变换, 时间序列重构