公交到站时间预测模型的研究

发布时间：2018-02-10 06:30

本文关键词： LSTM网络公交到站时间预测深度学习循环神经网络迁移学习　出处：《青岛大学》2017年硕士论文　论文类型：学位论文

【摘要】：公交到站时间预测是城市居民出行服务的需要,是广大乘客最关心的交通信息之一,它的预测准确度直接影响着乘客对公交系统的满意度,预测准确度的提高可以节约广大乘客的等待时间,从而提高出行效率,进而提升公交系统的服务水平,吸引更多乘客乘坐公交车,有利于解决交通拥堵问题。本文针对公交到站时间预测问题进行了研究,公交到站时间既与当前到站时间有关,又与历史同期数据相关,基于此本文根据深度学习中的LSTM网络模型在时间序列处理方面的优点,构建了基于LSTM网络的公交到站时间预测模型,并通过实验的方法验证了此模型的有效性与可实施性。本文的主要工作如下:1.将现有多种公交到站时间预测模型进行介绍。详细介绍了深度学习的相关内容和循环神经网络与LSTM网络的内容和结构,分析并总结影响公交到站时间的静态、动态因素。2.建立了基于LSTM的公交到站时间预测模型。以青岛市公交到站时间预测为例,在对数据进行了缺失值及归一化处理的基础上,构建了基于LSTM的公交到站时间预测模型,并对青岛市公交车到站时间进行预测,结果表明了此预测模型的可行性与有效性。3.优化模型。为了提高模型的预测精度,对模型进行了优化,最后面对特定的训练数据样本比较少的问题,通过迁移学习的方式对数据集进行扩充,来提高网络模型的性能,进而提高预测准确度。
[Abstract]:Bus arrival time prediction is the need of urban residents travel service, is one of the most concerned traffic information of the majority of passengers, its prediction accuracy directly affects the passenger satisfaction to the bus system. The improvement of prediction accuracy can save the waiting time of the vast number of passengers, thus improve the travel efficiency, and then improve the service level of the public transport system, and attract more passengers to take the bus. In this paper, the problem of bus arrival time prediction is studied. The bus arrival time is not only related to the current arrival time, but also related to the historical data. Based on this, according to the advantages of LSTM network model in deep learning in time series processing, a bus arrival time prediction model based on LSTM network is constructed. The main work of this paper is as follows: 1. The existing bus arrival time prediction models are introduced. The related contents of the in-depth learning and follow are introduced in detail. The content and structure of loop neural network and LSTM network, The static and dynamic factors affecting bus arrival time are analyzed and summarized. 2. A bus arrival time prediction model based on LSTM is established. Taking Qingdao bus arrival time prediction as an example, the missing values and normalized data are processed. The bus arrival time prediction model based on LSTM is constructed, and the bus arrival time in Qingdao is forecasted. The results show that the prediction model is feasible and effective. In order to improve the performance of the network model and improve the prediction accuracy, the model is optimized, and the data set is expanded by migration learning for the problem of fewer training data samples.
【学位授予单位】：青岛大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18;U491.17

【参考文献】