随着智能交通系统的快速发展和广泛应用，道路交通流量分析和处理的研究越来越多。作为交通运输是否进入信息时代的标志，智能交通系统将成为我国交通运输体系的发展方向。交通流预测是智能交通的重要组成部分，预测未来时段交通流状况对缓解交通拥堵，有效利用道路资源有着重要的意义。
交通流预测的研究模型有很多种，如：神经网络模型、多元线性回归模型、时间序列模型、历史趋势模型、kalman滤波模型等。而本文则着重研究kalman滤波在交通流预测中的应用。
本文研究了交通流的混沌特性，对交通流的可预测性进行判别。结合相空间重构理论建立相空间中的Kalman滤波交通流预测模型。本文选择了C-C算法进行相空间重构参数的计算。此外，本文为了改善相空间中的Kalman滤波模型预测效果，提出了利用两周中相对应时间的交通流差值或者比值代替原始数据，建立相空间差值回归预测模型和相空间比值回归预测模型。通过实际交通流的实验仿真，计算模型性能指标，并进行比较分析。本文将所建立预测模型与基于BP神经网络的交通流预测模型作对比，研究表明本文算法性能指标要优于BP神经网络预测模型。
本文还通过增加原始数据的方法，建立多周数据相空间比值回归模型，并与单周数据相空间比值回归模型进行性能对比分析，证明了其预测的优势之处。
最后，本文研究了多点数据融合在交通流预测中的应用。本文将数据融合理论应用在相空间的Kalman滤波交通流预测模型中，并对实际数据进行仿真验证，将单点数据相空间的Kalman滤波预测模型和多点数据融合的相空间Kalman滤波预测模型进行性能对比分析，表明多点数据融合理论在交通流预测中有较好的应用效果。

With the rapid development and wide application of intelligent transportation system, there are more and more road traffic flow analysis and processing. As the sign of transport going into the information age, intelligent transportation system will be the direction of development of China's transportation system. The traffic flow forecast is an important part of the intelligent Transportation system. It is very significant that predicting traffic flow conditions accurately for the future periods to alleviate traffic congestion and effectively use of road resources.
There are a lot of traffic flow forecasting models. For example, multiple linear regression model, time series model, neural network model, the historical trend model, Kalman filter model and so on. And this paper is focused on Kalman filtering in traffic flow forecast.
This paper deals with chaotic characteristics of the traffic flow and distinguishes the predictability of traffic flow. Here it establishes Kalman filter traffic flow forecasting model in the phase space with the combination of phase space reconstruction theory. And I choose C-C algorithm to calculate of the phase space reconstruction parameters. Besides, in this paper in order to improve the prediction effect of Kalman filter model in the phase space, it brings out that use two weeks corresponding to the time difference or the ratio of the traffic flow instead of the original data to establish phase space difference value regression prediction model and phase space ratio regression prediction model. Then by the experimental simulation of the actual traffic flow, it calculates the model performance index and carries out comparative analysis.
This paper compares the prediction model that we established to the traffic flow forecasting model based on BP neural network, and the study shows that the algorithm performance is better than the BP neural network model.
Besides, this paper establishes multi-weeks’ data phase space ratio regression model by the way of increasing the raw data, and compares to the single-week phase space ratio regression model to analyse its performance. It proves that it is better than others.
Finally, this paper studies the multi-data fusion in traffic flow forecast. In this paper, it applies the data fusion theory to the phase space of the Kalman filter traffic flow forecasting model and does some simulations on the actual data. Then it compares the single point of data the phase space of the Kalman filter prediction model to multi point of data the phase space of the Kalman filter prediction model. It shows that multi point of data the phase space of the Kalman filter prediction model is better in the traffic flow forecast.