乘客绩效导向的快速公交系统评估研究

发布时间：2019-06-14 02:13

【摘要】：快速公交系统(BRT)是一种解决城市交通拥堵、环境污染、资源浪费等问题的有效手段。但BRT系统在我国的发展时间较短,除了少有的几个城市,许多城市发展BRT处于示范引导作用。因此,本文从乘客的角度,选取BRT乘客绩效指标,采用改进了的评估模型,研究乘客绩效导向的BRT系统的评估,通过城市间的横向对比,实现公共财政投入到BRT系统的绩效监管,既可为政府指出公共财政投资BRT系统建设、运营的着力点,也可为已建成BRT系统的城市提供高效评估的方法。本文回顾了国内外关于BRT系统绩效评估以及模糊神经网络集成方法等方面的相关研究。在此基础上,对BRT系统的构成与功能定位,基于不同主体的绩效展开阐述。对绩效指标进行分析整合出乘客最关注的BRT的三个准则7个指标,构成本文的评估指标体系。同时,基于模糊评价法与神经网络的优缺点,提出采用两者集成的思想对评估模型进行改进。然后,提出模糊BP神经网络的评估步骤与数据处理方法,以及虚拟样本生成技术解决训练样本量不足的方法。最后,以广州、杭州、常州和济南为的绩效指标数据为基础,基础数据通过数据归一化处理、虚拟样本生成技术、模糊评价法等处理,得到神经网络的训练样本。最后通过样本测试,完成验证,得出相关结论。最后,本文所选的指标能够反映乘客的绩效关注点,改进的模糊BP神经网络模型能较为客观、高效完成绩效评估的计算。所采用虚拟样本生成技术拓展样本,既能保证原始样本的特征,并提高神经网络的泛化能力,在其他城市同类项目的绩效评估中具有可推广性。论文创新点包括以下两个方面:首先,模糊BP神经网络模型是一种基于模糊评价法与神经网络集成的改进方法,本文对模型的改进,并利用其它城市的数据进行训练,并通过测试,验证模型的可用性。其次,对于神经网络性能而言,过大或过少的训练样本均会影响神经网络的泛化能力和适应性。本文提出基于扰动思想的虚拟样本生成技术拓展样本,不仅能够提神经网络的泛化能力,而且能够有效避免陷入局部最小值。
[Abstract]:Bus rapid transit system (BRT) is an effective means to solve the problems of urban traffic congestion, environmental pollution, waste of resources and so on. However, the development time of BRT system in China is relatively short. In addition to a few cities, many cities are in the role of demonstration and guidance of BRT development. Therefore, from the point of view of passengers, this paper selects BRT passenger performance index, adopts the improved evaluation model, studies the evaluation of passenger performance-oriented BRT system, and realizes the performance supervision of public financial investment into BRT system through the horizontal comparison between cities, which can not only point out the focus of public finance investment BRT system construction and operation, but also provide an efficient evaluation method for cities that have built BRT system. In this paper, the research on performance evaluation of BRT system and fuzzy neural network integration method at home and abroad are reviewed. On this basis, the composition and functional positioning of BRT system are described based on the performance of different subjects. The analysis of the performance indicators integrates the three criteria and seven indicators of BRT, which are of the greatest concern to passengers, and constitutes the evaluation index system of this paper. At the same time, based on the advantages and disadvantages of fuzzy evaluation method and neural network, the idea of integration of fuzzy evaluation method and neural network is proposed to improve the evaluation model. Then, the evaluation steps and data processing methods of fuzzy BP neural network are put forward, and the method of virtual sample generation technology to solve the shortage of training samples is put forward. Finally, based on the performance index data of Guangzhou, Hangzhou, Changzhou and Jinan, the basic data are processed by data normalization, virtual sample generation technology, fuzzy evaluation method and so on, and the training samples of neural network are obtained. Finally, through the sample test, complete the verification, draw the relevant conclusions. Finally, the indicators selected in this paper can reflect the performance concerns of passengers, and the improved fuzzy BP neural network model can complete the calculation of performance evaluation more objectively and efficiently. The virtual sample generation technology can not only ensure the characteristics of the original sample, but also improve the generalization ability of the neural network, which can be extended in the performance evaluation of the same kind of projects in other cities. The innovation of this paper includes the following two aspects: firstly, the fuzzy BP neural network model is an improved method based on the integration of fuzzy evaluation method and neural network. In this paper, the model is improved and trained with the data of other cities, and the availability of the model is verified by testing. Secondly, for the performance of neural network, too large or too few training samples will affect the generalization ability and adaptability of neural network. In this paper, a virtual sample generation technique based on disturbance idea is proposed to expand the sample, which can not only improve the generalization ability of neural network, but also effectively avoid falling into the local minimum.
【学位授予单位】：长安大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U491.17

【参考文献】