基于优化极限学习机的公交行程时间预测方法

Bus Travel Time Prediction Based on Extreme Learning Machine Optimized by Artificial Bee Colony Algorithm

为提高城市公交车辆行程时间的预测精度,在分析历史数据和交通流变化特性基础上,提出了一种基于人工蜂群优化的极限学习机的组合预测模型(artificial bee colony-extreme learning machine,ABC-ELM)。首先,基于GPS等数据提取站间距离、时间周期及天气情况等动静态特征因素;其次,推算出公交车辆的站点停靠时间;接着,将人工蜂群优化算法(artificial bee colony algorithm,ABC)嵌入到传统极限学习机算法(extreme learning machine,ELM)中,以解决其在行程时间预测中收敛速度慢、初始权值和阈值选择困难的问题;最后,基于ABC-ELM算法预测公交车辆在目标路段的行程时间。以深圳市620路公交车的真实运营数据为基础进行方法验证。结果表明:与广泛采用的BP神经网络、SVM和ELM相比,本文方法在不同道路环境中均能保持较低的预测误差(RMSE误差:高峰/平峰为11.91/8.72,工作日/非工作日为11.46/9.54,晴天/雨天为10.83/12.31;决定系数R 2:高峰/平峰为0.87/0.92,工作日/非工作日为0.83/0.88,晴天/雨天为0.89/0.85),鲁棒性较强,更适用于复杂城市道路环境中的干线公交车辆的行程时间预测。

In order to improve the prediction accuracy of bus travel time,a combined prediction model based on artificial bee colony optimization and extreme learning machine(artificial bee colony-extreme learning machine,ABC-ELM)are proposed after analyzing historical data and the characteristics of traffic flow.First,dynamic and static characteristics like distance between stations,time period and weather conditions are extracted by using IC card and GPS data;after that the dwell time of the station is calculated.Then,the artificial bee colony optimization algorithm(ABC)is embedded in the traditional extreme learning machine algorithm(ELM)to solve the problem of slow convergence speed and difficulty in selecting initial weights and thresholds ELM in bus travel time prediction.Finally,the travel time of the bus on target road section is predicted by using the ABC-ELM algorithm.The model is verified based on the real operating data of Shenzhen Bus 620.The results show that,compared with the widely used BP neural network,SVM and ELM,the method proposed in this paper can maintain lower prediction errors in different road environments and has strong robustness(the RMSE error in peak/off-peak hour is 11.91/8.72,in workday/non-work day is 11.46/9.54,in sunny/rainy day is 10.83/12.31;the coefficient of determination R 2 in peak/off-peak hour is 0.87/0.92 in workday/non-work day is 0.83/0.88,in sunny/rainy day is 0.89/0.85),which makes it more suitable for travel time prediction in complex urban road environment and for main line bus.