基于ELM的成山头短时船舶交通流预测研究

发布时间：2018-01-29 10:47

本文关键词： 短时船舶交通流预测神经网络超限学习机　出处：《大连海事大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着我国海运事业的迅速发展,我国沿海已经成为世界上水上交通最繁忙的区域之一。航经我国沿海水域的船舶数量逐渐增多,海上交通事故也有增加的趋势,带来交通安全、海洋环境污染等一系列问题。短时交通流预测是交通流预测的重点研究内容之一,及时、准确的短时船舶交通流预测信息是保障船舶通航安全、航道畅通、交通有效运行的关键。本文在总结现有道路短时交通流预测和船舶交通流预测模型的基础上,针对现有短时交通流预测算法精度低、收敛慢、性能不稳定等问题,对短时船舶交通流预测问题进行了研究,并引进在道路短时交通流预测方面应用广泛的超限学习机(Extreme Learning Machine,ELM)对成山头分道通航水域的短时船舶交通流进行预测。本文首先简单介绍了交通流理论基础知识和短时船舶交通流预测的基本概念,并对船舶交通流特征、交通流预测的基本特征和基本原则、预测方法模型要求、现有预测方法及各自的与缺点进行了分析,并给出了短时船舶交通流预测模型的评价指标。其次,应用数据获取软件从海事局AIS网站获取成山头分道通航水域的船舶信息,并对数据进行筛选。为减少数据波动对基于ELM神经网络短时船舶交通流预测模型的影响,对采集到的数据进行了标准归一化处理,提高模型的预测精度。然后,应用MATLAB这一数学工具建立了基于ELM神经网络的成山头短时船舶交通流预测模型,并对编程实现进行了介绍。最后,分别采用小波神经网络的时间序列预测模型、BP神经网络非线性预测模型和ELM神经网络预测模型来对成山头分道通航警戒区进行短时交通流预测。研究结果表明,在一定误差范围内,基于小波神经网络的时间序列预测模型、BP神经网络非线性预测模型和基于ELM神经网络预测模型均能较好的预测成山头分道通航水域的短时船舶交通流量。对比分析上述3种预测模型的预测结果和误差,分析结果表明ELM神经网络模型能够有效避免传统神经网络的固有缺陷,应用ELM神经网络对短时船舶交通流预测的误差更小、运算时间更快、精度更高。更能满足短时船舶交通流预测的要求。
[Abstract]:With the rapid development of marine transportation in our country, the coastal area of our country has become one of the busiest water transportation areas in the world. The number of ships passing through the coastal waters of our country is increasing gradually, and the number of maritime traffic accidents is also increasing. Short-term traffic flow forecasting is one of the key research contents of traffic flow forecasting. Timely and accurate short-term ship traffic flow forecasting information is to ensure the navigation safety of ships. On the basis of summarizing the existing short-term traffic flow forecasting and ship traffic flow forecasting models, this paper aims at the low precision and slow convergence of the existing short-term traffic flow forecasting algorithm. The prediction of short-time ship traffic flow is studied in this paper. It also introduces extreme Learning Machine, which is widely used in road short-term traffic flow prediction. In this paper, the basic knowledge of traffic flow theory and the basic concept of short-term ship traffic flow prediction are introduced. The characteristics of ship traffic flow, the basic characteristics and principles of traffic flow prediction, the requirements of forecasting method model, the existing forecasting methods and their respective shortcomings are analyzed. The evaluation index of short-term ship traffic flow prediction model is given. Secondly, the data acquisition software is used to obtain the ship information from the AIS website of MSA. In order to reduce the influence of data fluctuation on short-term ship traffic flow prediction model based on ELM neural network, the collected data are normalized. The prediction accuracy of the model is improved. Then, the prediction model of short-time ship traffic flow based on ELM neural network is established by using MATLAB, and the programming implementation is introduced. Time series prediction models based on wavelet neural network are used respectively. The nonlinear prediction model of BP neural network and the ELM neural network prediction model are used to predict the traffic flow in the navigational warning area of Chengshan Mountain head. The results show that the traffic flow is within a certain error range. Time series prediction model based on wavelet neural network. Both the BP neural network nonlinear prediction model and the ELM neural network prediction model can be used to predict the short-time ship traffic flow in the Shantou and navigable waters. The prediction results of the above three forecasting models are compared and analyzed. And errors. The analysis results show that the ELM neural network model can effectively avoid the inherent defects of the traditional neural network, the application of ELM neural network to short-term ship traffic flow prediction error is smaller, faster operation time. The accuracy is higher and can meet the demand of short-time ship traffic flow forecast.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U692

【参考文献】