基于出行模式和神经网络的地铁短时客流预测方法研究

发布时间：2018-01-12 22:19

本文关键词：基于出行模式和神经网络的地铁短时客流预测方法研究　出处：《吉林大学》2017年硕士论文　论文类型：学位论文

【摘要】：建设轨道交通系统是缓解交通压力的有效途径之一,对地铁短时客流的准确预测可以为地铁车次的智能调度、站点限流与客流疏散方案的制定提供依据。本文针对短时客流具有非线性、时变性的特点,选择神经网络作为预测模型,并提出将通勤因素与短时客流预测结合。由于神经网络的性能很大程度上依赖于模型初始参数的设置,因此本文提出一种改进的萤火虫算法用于优化神经网络的初始参数。本文的主要工作如下:(1)基于上海市地铁一卡通数据对地铁客流进行特征分析:针对周内客流特征存在差异,使用层次聚类法对客流聚类,并借鉴上海市综合交通年度报告分析周五及长假前一日客流的特征,对聚类结果进一步细化;计算待预测时间片客流量和历史客流序列之间的Spearman相关系数;根据乘客出行链的相关理论设计了10种出行模式,在此基础上提出本文对“通勤”的定义,利用Hadoop平台编程计算各站点在各时间片内的通勤乘次,并说明本文从一卡通数据中分离出的通勤客流在一段时间内具有时空稳定性。(2)基于上海市气象局、环保局提供的相关数据分析降雨量和空气质量指数对短时客流的影响。(3)将时间相关性较高的客流序列作为输入,比较BP神经网络和Elman神经网络的预测性能,进一步缩小客流序列的最优输入维数所处区间,并选定更适应时变性的Elman神经网络作为预测模型;将通勤、天气因素与短时客流预测结合,验证本文提出的通勤因素能大幅提高预测精度,并选定性能最好的输入组合作为预测模型的输入。(4)介绍元启发式优化算法,详细分析了萤火虫算法(FA)的原理、流程和优缺点,并针对其存在的缺点提出改进:引入混沌机制和“鲶鱼效应”提高算法的全局搜索能力;引入Levy飞行提高算法的局部探索能力;对每只萤火虫个体采用自适应步长策略以提高算法的寻优精度。通过比较不同优化算法的收敛速度和寻优精度,验证了改进的FA算法的有效性;通过比较不同模型的预测性能,验证了基于改进的FA算法优化后的预测模型的有效性。
[Abstract]:The construction of rail transit system is one of the effective ways to relieve traffic pressure. The accurate prediction of subway short-term passenger flow can be used for the intelligent dispatching of subway train. In view of the nonlinear and time-varying characteristics of short-term passenger flow, the neural network is selected as the prediction model. It is proposed that the commuting factor be combined with the short-term passenger flow prediction, because the performance of the neural network depends on the setting of the initial parameters of the model to a great extent. Therefore, an improved firefly algorithm is proposed to optimize the initial parameters of neural network. The main work of this paper is as follows: 1). Based on the data of Shanghai Metro Card, this paper analyzes the characteristics of subway passenger flow: there are differences in the characteristics of the passenger flow during the week. The hierarchical clustering method is used to cluster the passenger flow, and the characteristics of the passenger flow on Friday and 1st before the long holiday are analyzed by using the annual comprehensive traffic report of Shanghai, and the clustering results are further refined. The Spearman correlation coefficient between the passenger flow and the historical passenger flow sequence was calculated. According to the related theory of passenger travel chain, 10 travel modes are designed. On the basis of this, the definition of "commuting" is put forward in this paper, and the commuting times of each station in each time slice are calculated by programming with Hadoop platform. It also shows that the commuter flow separated from the card data in this paper has temporal and spatial stability for a period of time.) based on the Shanghai Meteorological Bureau. Relevant data provided by EPA to analyze the effect of rainfall and air quality index on short-term passenger flow. The prediction performance of BP neural network and Elman neural network is compared to further reduce the interval of optimal input dimension of passenger flow sequence. Elman neural network, which is more suitable for time-varying, is chosen as the prediction model. By combining commuting, weather factors and short-term passenger flow forecasting, it is verified that the commuting factors proposed in this paper can greatly improve the prediction accuracy. The best performance input combination is selected as the input. 4) the heuristic optimization algorithm is introduced, and the principle, flow chart, advantages and disadvantages of the firefly algorithm are analyzed in detail. In view of its shortcomings, some improvements are put forward: the introduction of chaos mechanism and the "catfish effect" to improve the global search ability of the algorithm; Levy flight is introduced to improve the local exploration ability of the algorithm. For each individual, adaptive step size strategy is used to improve the accuracy of the algorithm. The effectiveness of the improved FA algorithm is verified by comparing the convergence speed and the optimization accuracy of different optimization algorithms. The effectiveness of the optimized prediction model based on the improved FA algorithm is verified by comparing the prediction performance of different models.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U293.13;TP183

【相似文献】