基于智能理论的交通流量组合预测模型研究

发布时间：2018-07-31 13:06

【摘要】：随着社会经济的发展,科学技术的进步,人们交通出行的需求也不断上升,机动车的数量更是与日俱增。机动车数量的高速增长引发交通拥堵、道路安全、环境污染、能源消耗等一系列问题日益严峻。大量实践经验表明,智能交通系统是改善城市交通状况,提高公共交通的服务质量最为有效的手段,不论从生态环境角度还是从社会经济角度来说均具有重要的意义。交通流量预测是智能交通实现的前提和关键,如何准确地预测交通量信息已经成为学者们研究的热点。交通量受到多种随机干扰因素的影响,具有一定的随机性和动态性,因此交通量预测十分复杂,虽然很多学者已经对此展开深入研究,取得了一定的成果,但是有关交通流预测理论的研究至今没有形成较为成熟完备的理论体系,到底选取哪些预测模型,以及使用哪些方法进行改进才合理仍然是值得研究的问题。本文以智能预测理论为前提,以提高交通车流量的预测精度为目的,深入探讨不同的智能方法和模型之间的有效组合形式,主要涉及以下几个方面的研究工作。1.以灰色系统理论中的DGM(1,1)为基础,提出了DGM-GRNN和DGM-SVM两种组合预测模型。灰色系统GM是在小样本,贫信息的情况下经常使用的一种预测模型,现有的灰色组合模型多采用GM(1,1)和BP神经网络进行组合的方式。由于GM(1,1)没有考虑未来可能加入的干扰因素对系统带来的影响,因此不适合中长期预测,而DGM(1,1)模型采用离散形式对常规GM(1,1)建模进行改进,弥补了传统GM(1,1)模型的缺陷。另外,BP网络易陷入局部极小值,并且计算结果随机性较强,因此本文提出了两种残差修正的DGM预测模型。首先使用DGM(1,1)模型对原始数据序列进行预测,然后分别使用GRNN和SVM模型对尾段残差序列进行训练并获取残差的预测序列,最后合成最终的预测结果。在实际交通流预测实验中,通过与传统的GM(1,1)和DGM(1,1)进行对比,本文提出的组合模型在预测精度方面有了明显提高,验证了组合模型的有效性。2.提出了基于粒子群算法优化的支持向量机SVM预测模型。SVM以统计学理论和结构风险最小化原理为基础,能够在小样本学习环境下进行分类和预测,是智能模型研究的热点。本文深入探讨了SVM中不同参数对系统预测的影响,针对SVM预测过程过参数设置不准确的缺点,分别以LSSVM和SMOSVM为基础,提出了两种基于粒子群优化的PSO-LSSVM和SMOSVM交通量预测模型。首先利用PSO算法对惩罚参数C和核宽度。进行优化,确定最优的C和σ,然后分别使用SMOSVM和LSSVM通过交叉验证的手段进行交通流的预测。通过与试凑法和网格法优化的模型进行对比分析实验,本文提出的模型具有更好的预测性能,表明其能够有效的预测实时交通流量的变化趋势。3.提出了基于遗传算法优化的极限学习机预测模型GA-ELM。BP神经网络在训练调整各种参数的过程中,消耗了大量的时间,而极限学习机ELM在大大缩短网络训练时间的同时并没有降低网络的收敛能力。ELM随机设置输入层与隐含层的连接权值和阈值,在训练过程中又不再重新调整这些参数的值,使得隐含层节点的有效性有待提高。针对这一问题,在网络结构确定的条件下,本文将遗传算法用于极限学习机权重和阈值的选择过程,通过目标函数的使用,比随机赋值获得了更优的权重和阈值,也使得隐含层和输出层之间的连接权值矩阵更为合理。通过与BP模型、GA-BP模型以及标准的ELM模型实验对比,进一步验证了本文提出的模型在预测精度和运行时间上的优势。4.提出了时间序列、神经网络和灰色理论定权和不定权组合的交通流预测模型。目前,统计理论、非线性理论和智能理论在交通流中都有不同的应用,各种单一模型具有一定的优点,但又存在一定的片面性。为了综合利用各种预测模型所提供的信息,取长补短,本文以交通流预测常用的GM(1,1)、ARIMA和GRNN模型为基础,提出了一种固定权系数和一种变权系数的组合预测模型。在提出了准预测绝对误差的定义之后,分别建立了以相对误差为依据的定权组合模型和Elman变权组合模型。实验结果表明,组合模型特别是变权系数组合模型在相对误差、均方根误差以及均等系数方面均优于单一模型或两种模型组合,从而验证了其有效性与可行性。5.综合对比分析各种组合预测模型。对不同的组合预测模型使用相同的交通流量数据进行实验,分析和探讨各种预测模型的适用性。论文的创新之处主要体现在利用智能模型的有效组合进行交通流的预测。分别使用GRNN和SVM对DGM模型进行残差修正,进一步提高了模型的预测精度；使用仿生学的粒子群算法和遗传算法分别应用于SVM和ELM,通过对模型参数的有效优化,模型取得了更优的预测效果；提出了基于GM、ARIMA和GRNN的定权组合和Elman变权组合,实验结果表明组合模型能够获得更好的性能评价指标。综上所述,本文的研究结果具有一定的理论和应用价值,为更准确地进行交通流预测提供新思路和新途径。
[Abstract]:With the development of the social economy and the progress of science and technology, the demand for traffic travel is increasing, the number of motor vehicles is increasing. The rapid growth of motor vehicles leads to traffic congestion, road safety, environmental pollution, energy consumption and so on. A large number of practical experiences show that the intelligent transportation system is changed. The traffic condition of the city and the most effective means to improve the service quality of the public transportation are of great significance both in the ecological environment and from the socioeconomic point of view. Traffic flow forecast is the precondition and key of the realization of intelligent traffic. How to accurately predict the traffic information has become a hot spot of research by the scholars. The quantity is influenced by a variety of random interference factors and has a certain randomness and dynamics, so the traffic volume prediction is very complicated. Although many scholars have studied it in depth and achieved some results, the research on traffic flow prediction theory has not formed a more mature and complete theoretical system, which is to be selected in the end. Some prediction models and which methods to be improved is still a problem that is still worth studying. In this paper, based on the intelligent prediction theory, in order to improve the prediction accuracy of traffic traffic, the effective combination form between different intelligent methods and models is discussed, and the research work of.1. is mainly involved in the following aspects. On the basis of DGM (1,1) in grey system theory, two combination forecasting models of DGM-GRNN and DGM-SVM are proposed. The grey system GM is a prediction model often used in the case of small sample and poor information. The existing grey combination models are mostly composed of GM (1,1) and BP neural network. Because GM (1,1) does not consider the possible future The influence of the interfering factors on the system is not suitable for medium and long term prediction, and the DGM (1,1) model improves the conventional GM (1,1) model in discrete form, and makes up for the defects of the traditional GM (1,1) model. In addition, the BP network is easy to fall into the local minimum, and the result of the calculation is more random. So two kinds of residual correction are proposed in this paper. The DGM model is used to predict the original data sequence using the DGM (1,1) model. Then, the GRNN and SVM models are used to train the residual sequence of the tail section and obtain the prediction sequence of the residual difference. Finally, the final prediction results are synthesized. In the actual traffic flow prediction experiment, the comparison with the traditional GM (1,1) and DGM (1,1) is made. The combined model proposed in this paper has improved the accuracy of prediction, and validates the effectiveness of the combined model.2.. The support vector machine SVM prediction model based on particle swarm optimization (PSO) optimization is based on the statistical theory and the principle of structural risk minimization, which can be classified and predicted in a small sample learning environment. It is an intelligent model, which is an intelligent model. In this paper, the influence of different parameters on the system prediction in SVM is discussed, and two kinds of PSO-LSSVM and SMOSVM traffic prediction models based on particle swarm optimization are proposed based on LSSVM and SMOSVM, which are based on LSSVM and SMOSVM. Firstly, the C and kernel width of penalty parameters are applied to the C and kernel width of the penalty parameters. Optimization, determine the optimal C and sigma, and then use SMOSVM and LSSVM to predict traffic flow through cross validation respectively. Through comparison and analysis experiments with the model optimized by the trial and the grid method, the proposed model has better prediction performance, which shows that it can effectively predict the change of real-time traffic flow. The potential.3. proposed the limit learning machine prediction model based on the genetic algorithm optimization GA-ELM.BP neural network consumed a lot of time in the process of training and adjusting the various parameters, while the limit learning machine ELM greatly shortened the network training time and did not reduce the convergence energy of the network.ELM randomly set the input layer and the hidden layer. In the training process, the values of these parameters are no longer adjusted in the training process, so the validity of the hidden layer nodes should be improved. Under the condition of the network structure, the genetic algorithm is applied to the selection of the weight and threshold of the limit learning machine, and the use of the target function is obtained. Better weights and thresholds make the connection weight matrix between the hidden layer and the output layer more reasonable. By comparing with the BP model, the GA-BP model and the standard ELM model experiment, it is further verified that the advantages of the model proposed in this paper are time series, neural network and grey theory in the prediction accuracy and running time. Traffic flow forecasting model of combination of fixed weight and indefinite right. At present, statistical theory, nonlinear theory and intelligent theory have different applications in traffic flow. All kinds of single models have certain advantages, but there are some one-sided. In order to make use of the information provided by all kinds of prediction models, this paper is based on traffic flow prediction. Based on the commonly used GM (1,1), ARIMA and GRNN models, a combination prediction model of the fixed weight coefficient and a variable weight coefficient is proposed. After the definition of the quasi prediction absolute error is put forward, the fixed weight combination model and the Elman variable weight combination model based on the relative error are established respectively. The experimental results show that the combined model is especially variable. The weight coefficient combination model is superior to the single model or two model combinations in relative error, root mean square error and equal coefficient, thus validates its effectiveness and feasibility.5. comprehensive comparison and analysis of various combination forecasting models. The innovation of this model is mainly embodied in the use of the effective combination of intelligent models to predict the traffic flow. The residual correction of the DGM model by using GRNN and SVM is used to further improve the prediction accuracy of the model. The bionics particle swarm optimization and genetic algorithm are used for SVM and ELM respectively. The model parameters are optimized effectively, and the model has better prediction effect. The combination of weight and Elman based on GM, ARIMA and GRNN is proposed. The experimental results show that the combined model can obtain better performance evaluation index. In summary, the results of this paper have a definite theoretical and application value for more accurate delivery. Flow prediction provides new ideas and new ways.
【学位授予单位】：东北师范大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：U491.14

【相似文献】