公交动态调度系统中的数据预测方法
发布时间：2018-12-24 16:31
【摘要】：智能公交系统因其具有高效的公交客运组织模式、快速灵活的应变能力、完善的乘客信息服务等优点越来越受到人们的关注,而动态调度作为系统中的一个核心环节,是影响公交运营系统运行成本、效率和服务水平的重要内容。目前的动态调度主要依靠调度员的经验,这种方式不稳定、可靠性差,造成调度不科学的原因之一是调度员缺乏数据依据,对未来的客流情况、路况信息以及公交车到站时间数据不了解。本文基于动态调度缺乏数据依据的现状,研究对调度有影响作用的客流数据、路况数据和到站时间数据的变化规律,建立预测模型,预测出客流数据、路况数据和车辆到站时间数据未来的变化情况,为动态调度提供数据支撑。主要进行了如下研究工作：首先,分析了影响公交车辆到站时间的因素,最终确定将站点客流和实时路况作为本文研究的两个因素。通过分析站点客流数据和实时路况数据的时空分布规律,提出了本文的数据预测思路。其次,利用站点客流数据和交通流速度数据本身存在的周期性和规律性,对传统时间序列ARMA模型进行了改进,建立了基于季节ARIMA的预测模型。采用实际采集的数据对站点客流和交通流速度预测模型进行验证,并与传统预测方法进行对比,本文提出的基于季节ARIMA的预测模型效果最优,对站点客流的预测MAPE为15.9%,对交通流速度的预测MAPE为6.84%,都在可接受范围内。再次,考虑到出行人和驾驶员有自已的行为习惯,导致交通信息模型存在一定的“自重复”,利用“自重复”特点建立了基于K近邻非参数回归预测模型,另外考察了模型中的关键参数回溯系数m和近邻个数K对预测效果的影响。在最优参数的模型下,对站点客流的预测MAPE为26.1%,对交通流速度的预测MAPE为22.5%。最后,在充分考虑公交车到站时间数据的周期性的同时,加入了动态调整部分,将站点客流对公交车站点停靠时间的影响和交通流速度对公交车路段行驶时间的影响进行了分析,最终建立了考虑前车数据和动态调整的到站时间预测模型。利用仿真模型产生的数据进行了验证,对30分钟后出发的班次的预测MAPE为11.5%,对60分钟后出发的班次的预测MAPE为10.6%,对90分钟后出发的班次的预测MAPE为5.6%,并与传统的基于GPS数据的预测方法进行对比,预测效果良好。并且对数据波动、权重变化和前车数据的影响三个因素进行了参数实验,找到了特定状态下的最优参数。
[Abstract]:Intelligent public transport system has attracted more and more attention because of its advantages such as efficient bus passenger transport organization mode, rapid and flexible response ability, perfect passenger information service, etc. Dynamic scheduling is a key link in the system. It is an important content that affects the operation cost, efficiency and service level of public transport operation system. The current dynamic scheduling mainly depends on the dispatcher's experience. This way is unstable, and the reliability is poor. One of the reasons for the unscientific scheduling is that the dispatcher lacks the data basis for the future passenger flow. Traffic information and bus arrival time data are unknown. Based on the lack of data basis for dynamic scheduling, this paper studies the changes of passenger flow data, road condition data and arrival time data, and establishes a prediction model to predict passenger flow data. The future changes of road condition data and vehicle arrival time data provide data support for dynamic scheduling. The main research work is as follows: firstly, the factors affecting the arrival time of public transport vehicles are analyzed, and the passenger flow and real-time traffic conditions are determined as two factors in this paper. By analyzing the temporal and spatial distribution of the station passenger flow data and the real time traffic condition data, this paper puts forward the data prediction thought of this paper. Secondly, using the periodicity and regularity of station passenger flow data and traffic flow velocity data, the traditional time series ARMA model is improved, and the prediction model based on seasonal ARIMA is established. The prediction model of passenger flow and traffic flow velocity is verified by the actual data collected, and compared with the traditional forecasting method. The forecasting model based on seasonal ARIMA is the best, and the MAPE of passenger flow prediction is 15.9. The MAPE for traffic flow velocity is 6. 84 and is within acceptable range. Thirdly, considering that travelers and drivers have their own behavior habits, which leads to the existence of "self-repetition" in the traffic information model, based on the characteristics of "self-repetition", a non-parametric regression prediction model based on K-nearest neighbor is established. In addition, the influence of the backtracking coefficient m and the number of nearest neighbors in the model on the prediction effect is investigated. Under the model of optimal parameters, the forecast MAPE of passenger flow at the station is 26.1and the MAPE of traffic flow velocity is 22.5. Finally, while fully considering the periodicity of bus arrival time data, the dynamic adjustment part is added. The influence of passenger flow on bus stop time and the influence of traffic flow speed on bus section travel time are analyzed. Finally, a prediction model of arrival time considering the data of the front car and dynamic adjustment is established. The simulation model is used to verify the results. The predicted MAPE of the flight after 30 minutes is 11.5, the MAPE of the departure after 60 minutes is 10.6, the predicted MAPE of the departure after 90 minutes is 5.6, the predicted MAPE of the departure after 30 minutes is 10.6, the predicted MAPE of the departure after 60 minutes is 10.6, the predicted MAPE of the departure after 90 minutes is 5.6. Compared with the traditional prediction method based on GPS data, the prediction effect is good. The parameters of data fluctuation, weight change and the influence of the front car data are tested, and the optimal parameters are found.
【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U491.17
本文编号：2390829
[Abstract]:Intelligent public transport system has attracted more and more attention because of its advantages such as efficient bus passenger transport organization mode, rapid and flexible response ability, perfect passenger information service, etc. Dynamic scheduling is a key link in the system. It is an important content that affects the operation cost, efficiency and service level of public transport operation system. The current dynamic scheduling mainly depends on the dispatcher's experience. This way is unstable, and the reliability is poor. One of the reasons for the unscientific scheduling is that the dispatcher lacks the data basis for the future passenger flow. Traffic information and bus arrival time data are unknown. Based on the lack of data basis for dynamic scheduling, this paper studies the changes of passenger flow data, road condition data and arrival time data, and establishes a prediction model to predict passenger flow data. The future changes of road condition data and vehicle arrival time data provide data support for dynamic scheduling. The main research work is as follows: firstly, the factors affecting the arrival time of public transport vehicles are analyzed, and the passenger flow and real-time traffic conditions are determined as two factors in this paper. By analyzing the temporal and spatial distribution of the station passenger flow data and the real time traffic condition data, this paper puts forward the data prediction thought of this paper. Secondly, using the periodicity and regularity of station passenger flow data and traffic flow velocity data, the traditional time series ARMA model is improved, and the prediction model based on seasonal ARIMA is established. The prediction model of passenger flow and traffic flow velocity is verified by the actual data collected, and compared with the traditional forecasting method. The forecasting model based on seasonal ARIMA is the best, and the MAPE of passenger flow prediction is 15.9. The MAPE for traffic flow velocity is 6. 84 and is within acceptable range. Thirdly, considering that travelers and drivers have their own behavior habits, which leads to the existence of "self-repetition" in the traffic information model, based on the characteristics of "self-repetition", a non-parametric regression prediction model based on K-nearest neighbor is established. In addition, the influence of the backtracking coefficient m and the number of nearest neighbors in the model on the prediction effect is investigated. Under the model of optimal parameters, the forecast MAPE of passenger flow at the station is 26.1and the MAPE of traffic flow velocity is 22.5. Finally, while fully considering the periodicity of bus arrival time data, the dynamic adjustment part is added. The influence of passenger flow on bus stop time and the influence of traffic flow speed on bus section travel time are analyzed. Finally, a prediction model of arrival time considering the data of the front car and dynamic adjustment is established. The simulation model is used to verify the results. The predicted MAPE of the flight after 30 minutes is 11.5, the MAPE of the departure after 60 minutes is 10.6, the predicted MAPE of the departure after 90 minutes is 5.6, the predicted MAPE of the departure after 30 minutes is 10.6, the predicted MAPE of the departure after 60 minutes is 10.6, the predicted MAPE of the departure after 90 minutes is 5.6. Compared with the traditional prediction method based on GPS data, the prediction effect is good. The parameters of data fluctuation, weight change and the influence of the front car data are tested, and the optimal parameters are found.
【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U491.17
【参考文献】
相关期刊论文 前1条
1 张春辉;宋瑞;孙杨;;基于卡尔曼滤波的公交站点短时客流预测[J];交通运输系统工程与信息;2011年04期
，本文编号：2390829
本文链接：https://www.wllwen.com/kejilunwen/jiaotonggongchenglunwen/2390829.html
