Clustering methods are analyzed in which Euclidean distance and network distance are used as a similarity measure respectively.The neighbor correlation between objects on a spatial network is discussed and a clustering algorithm is proposed for network objects with consideration of direction factors.The algorithm combines the two distances as the similarity measure of clustering by using the neighbor correlation.The analysis and experimental results indicate that the effectiveness of the proposed algorithm is better than those only using one measure.

A path planning method based on improved ant colony algorithm was proposed according to the path planning features in urban road traffic.This method enhances the descriptive ability of the real road traffic network to improve the effectiveness of path planning by analyzing and converting the traffic constraints.When the direction-heuristic information is introduced into the ant colony algorithm,there is enough initial search space to be held in order to improve the efficiency of path planning.Experimental results showed that the planning efficiency and effectiveness both increased evidently with application of the proposed method.

This paper introduces a data transmission approach based on high speed serial communication interface with preemptive task scheduling algorithm and Round-Robin and buffer pool.The method provides influent,synchronal and real-time downloading video stream in air.It is applied for large volume data transmission between air borne and ground and takes an important role in traffic surveillance.

Real-time accurate predicted traffic information is critical to intelligent traffic inducement and traffic management. We proposed a two-step learning algorithm: Genetic Gradient Algorithm (GGA) for RBF neural network. A genetic algorithm (GA) initially determines the parameters of the RBF network including not only the number and the locations of the selected centers but also the widths of Gaussian kernel functions in the hidden layer. Then a gradient descent algorithm is adopted to further adjust these parameters of the RBF network. Furthermore, we established an RBF network predicting model for traffic information prediction. The experimental results conducted on the real-time traffic speed information prediction in the city of Ningbo showed the good effects of our proposed methods.

In order to overcome single agent's knowledge incompleteness and improve the reliability of decision-making results,a multi-agent cooperative decision-making mechanism was proposed based on Dempster-Shafer theory.Formal description and definitions of the mechanism was given.The multi-agent cooperative decision-making was divided into two stages:learning and decision-making.The correct feedback was introduced into the multi-agent's learning procedure.According to the training cases and revision formulas,the conflicts among agents were resolved and the complexity of combinational computation was decreased,thus improving the work of Dragoni et al.This multi-agent cooperative decision-making mechanism is more reliable than the voting mechanism and the weighted majority algorithm.The experiment results show its effectiveness.

The main aspects of the multi-agent system(MAS) were agent-oriented, combined with formal program language Visual C++. A functional model of traffic system was constructed based on MAS, and on the basis of MFC library of Visual C++, the agent class that has the basic functions of cooperation, communication and autonomy is also designed.

The real-time and accurate predicted traffic information is critical to the intelligent traffic inducement and the traffic management. A two-step learning algorithm GGA (Genetic Gradient Algorithm) for radial basis function (RBF) neural network is proposed in this paper. A genetic algorithm (GA) initially determines the parameters of the RBF network including the number and locations of the selected centers and the widths of Gaussian kernel functions in the hidden layer. Then a gradient descent algorithm is adopted to further adjust these parameters of the RBF network. A smaller network with better generalization capability can thus be obtained. The experimental results of the real-time traffic information prediction in Ningbo city show good performance of the proposed method.

A predictive control algorithm based on high--order fuzzy internal model, which aimed at systems with dead time, is proposed. By introducing intelligent factor, the high--order fuzzy internal model of controlled plant can forecast output of process in time and modify model uncertainty so as to overcome the disadvantages of dead time. Running results certify its practical values in apply-- ing the algorithm to the distributed control systems of coal gasification reactors.

In real applications, there is often the need of estimating the intrinsic and extrinsic parameters of a camera directly from the images of natural scenes or working scenarios. When there are only several feature points can be determined, the results of current calibration methods are very unstable. In this paper, a geometrical method is presented for estimating a restricted camera. It can estimate f even when only one feature point is valid. Experiments show that it has good robustness and reliability, and it is tolerant to multiple error sources.

Simulation is an effective way to research traffic problem. At present, the traffic simulation softwares have two shortages. The one is that the softwares can not describe the height information because they are two-dimensional, the other is that the softwares can not handle the intelligent objects because they use OOP as the main method of simulation. This paper designs intelligent Agent and develops three-dimensional traffic simulation software with the help of OpenGL which is the 3d standard. The outcome of simulation shows that the technology of intelligent Agent improves the effect of traffic simulation.

An approach to detecting and tracking a moving object is presented. Foreground objects are segmented by using a per-pixel color-based hybrid Gaussian model as the background updating method, and tracked by template matching. A new method which improves the hybrid Gaussian model is presented, which allows the system to learn faster and more accurately and adapt effectively to the changing environments. Experiments show that the method is more robus than the state-of-the-art without sacrificing the real-time performance and well-suited for various chimates and lighting conditions.

Sparse Bayesian treatment is a state-of-the-art technique for regression and classification, combining excellent generalization properties with a sparse kernel representation. An approach to classify vehicle model in real time is presented in the paper. A per-pixel color-based mixture Gaussian model is used as the background updating method, thus vehicles are segmented. Sparse Bayesian classification is applied to recognize and classify vehicle models. Examples demonstrates Sparse Bayesian Classification achieves comparable recognition accuracy to the SVM, yet provides a full predictive distribution, and also requires substantially fewer kernel functions. Experimental results are efficient for classification.

Based on the theory of multiresolution analysis of wavelet transforms and fuzzy concepts, a new method called fuzzy wavelet support vector machines (FWSVM) was presented. The FWSVM consists of a set of fuzzy rules. Each rules corresponding to a sub-wavelet support vector machines (WSVM) with different resolution. Thus the sub-WSVM at different dilation value under these fuzzy rules is fully utilized to capture various essential components of the system. The role of the fuzzy set is to determine the contribution of the sub-WSVM to the output of the FWSVM. Through adjusting the parameters of membership functions, the model accuracy and the generalization capability of the FWSVM can be improved. Analysis of the experimental results proved that FWSVM could achieve greater accuracy than the standard SVM.

The system process was modeled and an inverse model controller using support vector machine regression (SVMR) was designed. The SVMR principle was briefly introduced. The SVMR was applied to the internal model control (IMC) problem, and the SVMR internal model was developed. An SVMR controller for internal model control problem was proposed under the inverse condition of control process. The control algorithm was applied to the reversible nonlinear system and greenhouse environment with unknown disturbance and was compared with neural networks IMC using simulation, and the results showed that the SVMR IMC had a simplified model and good control performance.

The generalities and specialties of rough sets (RS) and support vector machines (SVM) in knowledge representation and classification are analyzed. A minimum decision network combining RS with SVM in intelligent processing is investigated, and a kind of SVM system on RS is proposed for forecasting. Using RS theory on the advantage of dealing with great data and eliminating redundant information, the system reduced the training data of SVM, and overcame the disadvantage of great data and slow speed. Finally, the system is used to forecast short-term load. The experimental results proved that this approach could achieve greater forecasting accuracy and generalization capability than the BP neural network and standard SVM.

Based on PRAM computing model, the parallel processing for adaptive control algorithm based on fuzzy internal model, which aimed at nonlinear system, is discussed. By applying TSK-modeling scheme, the %m% processors identify fuzzy internal model on-line and design linear quadratic optimal (%H%-2-) is obtained in parallel by using the internal model control theory. The simulation experiment shows that the parallel processing method obtains sub-linear speedup, and the requirement for real-time identification of internal model on-line is basically satisfied.;

Based on the kernel method of support vector machines ( SVM) and wavelet frame theory, a new method called wavelet support vector machines (WSVM) is proposed. The method is applied to combination forecast, and an algorithm for constructing and training the forecast model of WSVM is presented. Simulation examples are also given to illustrate the effectiveness of the method comparing with the wavelet neural network and other methods.

在研究支持向量机(SVM)核方法和小波框架理论的基础上,提出了一种称为小波支持向量机(WaveletSupport Vector Machines,WSVM)的新的机器学习构造方法.该方法引入小波基函数构造SVM的核函数,得到了一种新的SVM模型,然后提出了此模型的结构设计和实现算法,最后给出了几种常用的小波核函数,并给出了理论证明.通过仿真实验,把该方法与小波神经网络、高斯核SVM相比较,得到了较好的实验结果,从而验证了该方法的正确性和有效性.

The paper brings forward that it is crucial for group that has knowledge inside many domains to make decision more reliably than single decision-maker. The author makes quantitative analysis of the semi-structural decision problem. Combining with the example of confirming of affecting factors and their weights in the procedure of the adjusting of industrial structure in agriculture, a practicable policy of group decision-making is described, which provide strong foundation for.

支持向量机(SVM)回归理论与神经网络等非线性回归理论相比具有许多独特的优点.讨论了建模中SVM核函数、损失函数的选取和容量控制等问题,并用实验加以验证.将SVM回归动态建模理论应用于非线性、时变、大时延温室环境温度变化的建模和预测,模型简单,预测效果好.