基于二次修正的短时行程时间预测模型

为了提高高速公路短时行程时间预测模型的精度和鲁棒性,同时缓解过度训练带来的过拟合效应,构建了基于小波神经网络和马尔可夫链的组合预测模型,并采用平均绝对误差、平均绝对百分比误差、均方根误差三个指标评价模型的预测效果.分析结果表明,在行程时间突变之后,组合预测模型较其他模型都有着更高的预测精度;同时,该模型在预测行程时间突变点时不存在延迟,说明其在高峰时段内有着更高的预测精度和更强的预测鲁棒性.

Short-term Travel Time Prediction Model Based on Secondary Correction

In order to increase both of the accuracy and the robustness for freeway short-term travel time prediction, as well as easing the over-fitting effect, which was brought by the extra training, a hybrid model was proposed on the basis of wavelet neural network and Markov chain. The forecasting performance of different models was examined by three measures, i.e., mean absolute error, mean absolute percentage error, root mean square error. The results show that the proposed hybrid model enjoys obvious superiority over the other models after the break point of travel time. Furthermore, no prediction delay was observed in the prediction of break point of travel time. In conclusion, the higher prediction accuracy and the better robustness were found in the hybrid model in peak hours.