基于BP神经网络的公交动态行程时间预测方法研究

公交行程时间的精确预测对于提升公交吸引力具有重要意义。本文基于公交车到离站的历史数据,综合考虑时间周期、站点、站间距离、天气等多个因素,建立了基于BP神经网络的公交车静态行程时间预测模型,以该模型为基础,采用动态迭代的方法,叠加多个站间行程时间预测结果,进一步构建了面向连续站点的公交车动态行程时间预测模型,实现对跨越多个站点的公交行程时间预测。以青岛市125路公交为例对算法进行测试。在模型的横向对比实验中,本模型预测结果的绝对误差均在50 s以内,平均绝对误差百分比(MAPE)为11.74%,均方根误差(RMSE)为23.15,R2的确定系数为0.905 1,SVM的MAPE、RMSE、R2 误差指标分别为:12.38%、38.33、0.743 6,LR对应的误差指标分别为:12.50%、25.59、0.884 1;在静态模型与动态模型的对比实验中,动态模型预测结果的MAPE为11.75%,RMSE为23.15,静态模型对应误差指标分别为:11.63%、26.74。研究结果表明,基于BP神经网络的公交动态行程时间预测模型比传统的静态预测方法具有更高的预测精度。

Research on Prediction Method of Bus Dynamic Travel Time Based on BP Neural Network

The accurate prediction of bus travel time is of great importance to the improvement of bus attraction. This paper establishes a model on the basis of BP neural network to forecast the bus static travel time between stations, using the historical data of the arrival time and departure time of buses and taking into account the time period, bus stations, distance between stations, weather and other areal characteristics. And then we construct a forecasting model for the time and space process of bus dynamic travel that superimposes multiple static travel time prediction results by the method of dynamic iteration. Through this model, we can achieve the time prediction of buses travelling across multiple stations. Taking bus line 125 in Qingdao as an example to test the algorithm, all of the absolute error of the prediction results of this model is within the 50 s and Mean Absolute Percentage Error(MAPE) is 11.74%, Root Mean Squared Error(RMSE) is 23.15, Coefficient of determination(R^2)is 0.905 1, while the MAPE, RMSE and R^2 of SVM error index is: 12.38%, 38.33, 0.743 6 respectively, the corresponding error index of LR is: 12.50%, 25.59, 0.884 1 respectively in the horizontal comparison test. In the comparison test between static prediction model and dynamic prediction model, the MAPE, RMSE of prediction result of dynamic model is 11.75%, 23.15, while the corresponding error index of static model is: 11.63%,26.74 respectively. It turns out that the bus dynamic travel time prediction model based on BP Neural Network can predict bus travel time better comparing to traditional static prediction methods.