Based On Manifold Learning Algorithms In Urban Traffic Control Theory And Applied Research

Intelligent control technique for urban traffic is a research focus in both traffic engineering field and control engineering filed. As the fast developing and applying of artificial intelligence, automatic control, computer science and communication technology, a variety of control means and methods emerged. New theories and productions appeared constantly. Their applications in practical projects and items show their tremendous powers and potentials.In real-world applications, observations represented as high-dimensional data or vectors can be modeled as samples lying on or close to a low-dimensional nonlinear manifold. Hence, data reduction especially nonlinear dimensionality reduction is an important tool of data mining, and the goal of dimension reduction is to find the low dimensional structure of the nonlinear manifold from the high dimensional data. The urban traffic system is nonlinear, stochastic, time-variant and uncertain complicated system with a lot of uncertain influencing factors. Traffic flow data base, as the research foundation, has some characters such as large amount, high dimensionality, strong randomicity, regulated and obvious nonlinear in a way. We suppose that there is a low-dimensional structure embedded in the detected traffic flow data set. With the precondition that the geometrical relationships and the distance measurements among data are kept unchanged, we map manifold corresponding to the original data in high dimension space into in low dimension space. So not only the data quantity is reduced in future relative calculation, but also disturbance of the noise is removed, and finally we can improve the efficiency of traffic signal control.We advance a method that reducing the dimension of data set with isometric mapping algorithm which is one of manifold learning algorithm. This paper focuses on isometric mapping algorithm and dimensional reduction. We analyze the superiorities and deficiencies of four main manifold learning algorithms, and decide to use the isometric mapping algorithm to diminish the dimension of detected traffic flow data set. The main contents of the dissertation are as follows.1. Through analyzing urban traffic system and traffic flow, we point out the key issue: the traffic flow data set of high-dimensionality and nonlinear, and present a method to resolve it: reduce the dimension with manifold learning algorithm.2. We make an overview of five main manifold learning algorithms, including MDS (Mutidimensional Scaling), Isomap (Isometric Mapping), LLE (Locally Linear Embedding), LE (Laplacian Eigenmap) and LTSA (Local Tangent Space Alignment). And we analyse their strengths and limitations, then advance an idea which using isometric mapping algorithm to diminish the dimension of traffic flow data.3. We bring forward a method to part time periods in urban traffic time of day control. First we reduce the data set's dimension with isometric mapping algorithm, and then cluster the low-dimension data set with K-means clustering algorithm, and finally we fix the time periods plan according to the clustering result. The result is as the same as outcomes of artificial method, hierarchical cluster analysis based on genetic algorithm, Artificial Immune partitioning analysis, but this method conquers the unreasonable of artificial method and improves the efficiency and avoids the shorts of later two means.4. We achieve the work of reducing dimension of traffic flow data set in urban road network. Considering that Isomap algorithm tend to quantities of data, and require different kind data separate not so insular in space, the paper test Isomap algorithm in road network and get the embedded dimension, realize the reduction of data set. And then we compare its result with outcomes of PCA (Principal Component Analysis).Isomap algorithm can diminish traffic flow data set's dimension and find out the embedded dimension. This feature will make the traffic signal actuated control much self-adapting, decrease the delay of vehicles and increase the traffic capacity. This thesis' study on intelligent traffic control is the foundation of latter research.