基于GA-LSSVR模型的路网短时交通流预测研究

目前,很多短时交通流预测方法仅利用某一路段历史数据的时间相关性或者道路上下游路段的时空相关性进行交通流预测,未充分考虑路网所有路段之间的时空相关性.提出了一种基于稀疏混合遗传算法优化的最小二乘支持向量回归(LSSVR)模型,并应用于路网短时交通流预测.该预测模型不仅可以自动优化LSSVR模型参数,而且可以从高维路网交通流数据中选择有助于交通流预测的变量子集.实验结果表明,与LSSVR模型相比,所提方法具有更好的预测能力;而且,少量时空变量被选择出来构建预测模型,极大减少了信息冗余,改进了模型可解释性.

Short-term Traffic Flow Forecasting of Road Network Based on GA-LSSVR Model

Currently, many short- term traffic flow forecasting methods just take into account either temporal correlation of historical data at the target road segment or spatiotemporal correlation between the upstream/downstream segments and the target one, thus ignoring complex spatiotemporal correlation from a more global viewpoint. Aiming at above problem, Least Squares Support Vector Regression (LSSVR) model optimized by a sparse hybrid Genetic Algorithm (GA) is put forward for short term traffic flow forecasting. This model not only automatically optimizes the involved parameters, but also selects from high-dimensional traffic data a subset of spatiotemporal variables contributing to traffic flow forecasting. The experimental results show that in comparison with LSSVR model, the proposed method can improve the performance of traffic flow forecasting. Moreover, only a few of spatiotemporal variables are selected by this method, not only reducing the information redundancy but also enhancing the interpretability of the resulting model.