融合多源数据预测高速公路站间旅行时间

为精确预测高速公路站间旅行时间，融合收费数据和微波车检数据开展预测. 首先，基于两种数据源的预测结果，采用决策级融合策略；然后，建立了权重分配预测模 型、BP神经网络预测模型；针对神经网络收敛速度慢，易陷入局部最优的缺陷，基于遗传 算法优化BP神经网络预测模型；最后，利用京哈高速公路北京段收费数据和微波检测器 数据对3 种融合模型进行了验证，对比工作日及非工作日2 种交通流状态下3 种模型的性 能指标.试验结果表明，基于遗传神经网络的融合模型相比其他2 种模型，预测精度及稳 定性均得到了较高的提升，相对误差控制在10%以内，能够更好地满足实际需求.

Highway Travel Time Prediction Based on Multi-source Data Fusion

In order to predict highway travel time accurately, toll collection data and microwave detection data are fused for travel time prediction. First, based on two prediction results, the decision level fusion strategy is determined. Then, the weight distribution model and back propagation neural network model are selected as the basic models. The next, because the neural network convergence is slow and easy to fall into local optimum, the Genetic algorithm is adopted to optimize the neural network model. Finally, on Beijing segment of Jingha Highway, suitable performance indices are proposed to compare the performance of three models in different traffic states, including weekday and weekend. The results show that the fusion model based on GA- BP neural network based on microwave detection data and toll data produced sufficient accuracy and stability. The Mean Relative Error (MPE) of all prediction periods are less than 10% for the normal and holiday traffic flow, which can better meet the practical requirements.