基于AML-BP算法的出行方式识别研究

发布时间：2018-06-26 01:53

本文选题：出行方式 + 方式识别　；参考：《郑州大学》2015年硕士论文

【摘要】：出行方式的识别是智能交通领域的一个新兴研究方向,主要通过对出行者出行数据如:速度、距离等的分析识别出行者的出行方式,用来对一定区域内出行习惯进行调查、分析人群出行特点,为市政决策提供帮助,同时,能够针对特定出行方式提供定制的道路引导方案。本文结合移动互联网和机器学习技术,设计并实现了一种基于AML-BP算法的出行方式识别方案。文章提出将城市中的步行、公共汽车、自行车、小汽车、电动摩托车五种目前主要出行方式作为识别目标,主要工作如下:第一,介绍了国内外出行方式识别的研究现状,给出了本文研究的整体思路,即:出行参数的采集、出行参数特征分析、出行方式识别算法的研究与改进、应用系统的实现;第二,开发了一款智能手机端应用,实现了对五种出行方式出行过程中的GPS和三轴加速度数据采集;第三,结合运动学、统计学知识,对各出行方式运动原理及采集到的出行数据进行分析,选择出对出行方式具有区别性的参数作为出行方式的特征;第四,选择BP神经网络算法作为出行方式识别算法,针对算法存在的弱点,提出了基于AML(Adaptive Momentum and Learning rate)的BP神经网络算法,并对AML-BP算法的识别效果进行了实验测试;最后,根据对出行方式识别的研究成果,通过Android平台技术、动态链接库技术、百度云服务技术等,实现了一款Android平台下的出行方式识别应用。根据选择的出行方式特征以及改进的BP算法,在与其他识别方案的对比中,本算法取得了较好的识别率。将出行方式识别算法应用于出行方式的Android应用中,在实际环境下对应用进行测试,应用运行稳定,达到了设计要求,在应用运行中能够对出行方式进行较准确的识别。
[Abstract]:The identification of travel mode is a new research direction in the field of intelligent transportation, mainly through the analysis of travel data such as speed, distance, etc., it can be used to investigate the travel habits in a certain area. The characteristics of crowd travel are analyzed to provide help for municipal decision making and to provide customized road guidance schemes for specific travel modes at the same time. Combined with mobile Internet and machine learning technology, this paper designs and implements a trip pattern recognition scheme based on AML-BP algorithm. In this paper, five main travel modes in the city, such as walking, bus, bicycle, car and electric motorcycle, are proposed. The main work is as follows: first, this paper introduces the current situation of the research on the identification of travel mode at home and abroad. The whole idea of this paper is given, that is, the acquisition of trip parameters, the characteristic analysis of trip parameters, the research and improvement of trip pattern recognition algorithm, the realization of application system. Secondly, a smart phone terminal application is developed. The data acquisition of GPS and triaxial acceleration in the travel process of five travel modes is realized. Thirdly, combined with kinematics and statistical knowledge, the principle of travel and the travel data collected are analyzed. Select the parameters that are different to the travel mode as the characteristics of the travel mode; fourth, select the BP neural network algorithm as the travel mode recognition algorithm, aiming at the weakness of the algorithm, A BP neural network algorithm based on Adaptive moment and Learning rate is proposed, and the recognition effect of AML-BP algorithm is tested experimentally. Finally, according to the research results of trip pattern recognition, through the technology of Android platform, dynamic link library technology, Baidu cloud service technology, the implementation of a Android platform under the travel mode identification application. According to the selected travel mode features and the improved BP algorithm, compared with other recognition schemes, the algorithm achieved a better recognition rate. The algorithm of trip pattern identification is applied to the Android application of trip mode, and the application is tested in the actual environment. The application runs stably and meets the design requirements, and it can accurately identify the trip mode in the application operation.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U495;TP391.4

【相似文献】