基于田口集成学习的最佳交通流量预测模型

Optimized traffic flow forecasting model with Taguchi ensemble learning approach

针对预测准确性对数据相关性和网络结构极为敏感的问题,采用田口方法选择集成学习的结构参数,在保证学习多样性的同时,提高选择过程的效率,保证预测的准确性。实验仿真计算结果表明,在多个路段的交通流量数据集上,该方法皆取得了满意的预测效果,模型具有较好的鲁棒性。将仿真结果与其它算法进行比较,比较结果表明,相比传统ARIMA、MLP模型,其预测精度分别提高了4.51%、11.23%。

To solve the problem that accuracy is extremely sensitive to data correlation and network structure,the forecasting model using the Taguchi method to select the ensemble learning structure parameters was designed.While ensuring the diversity of learning,the efficiency of the selection process was improved,and the accuracy of the forecasting was also guaranteed.Experimental results show that the proposed method can get satisfactory forecasting on the traffic data sets of multiple road segments,and the model has good robustness.Comparing the simulation results with the traditional ARIMA and MLP models,it is verified that the accuracy is improved by 4.51% and 11.23%respectively.