Application & Research On Traffic Flow Predication Based On LSSVM

Short-term traffic flow forecasting is the premise and key of intelligent traffic control and management, traffic flow state realization and real-time traffic guidance and is also the impersonality necessity. But so far, its results are unsatisfactory. Traditional forecasting methods based on accurate mathematical models which are computing hard, spending-time and needing too many historical data but less predication precision are no ideal. Therefore, studying artificial intelligence methods have some reality meanings. Based on comparative analyses of the existing short-term traffic flow models and research of characteristics of traffic flow, the thesis establishes the goal:the models established can remedy the shortcoming of the existing short-term traffic flow forecasting modelsSupport Vector Machine(SVM) is the new machine learning method based on statics learning theory(STL).it has the characteristic which contain the structure Risk Minimization of Statistical Learning Theory, strong predication emotion, optimum result of all data point. Compared to artificial network based on experience Risk Minimization, it has the more theory basement and application foreground. And due to the SVM model, it has the simple algorithm and less time spending and high predication precision. so it has been applied on traffic predication.This thesis based on Least Square Support Vector Machine theory use the principal "slide windows" to make a new online algorithm. The new algorithm can efficient dynamic renew the matrix and "pruning" the less efficient Support Vector. Using the MATLAB 7.0 to validate the new algorithm has feasibility. So it could be applied on Short-term traffic flow predication. Major jobs are as follow.First it can get the feature of SVM through the chapter 2.then we get the least Square method to mend the SVM algorithm in chapter 3. Base on these theory and algorithm, we use the classical method "pruning" to prove the efficiency and scientific of this method through computer simulation. Last using the matrix theory and apply the principle like "slide window" to renew the row/arrange. So the exchange algorithm could achieve the predication precision while satisfy the time-consume.In the last of this paper, we use the real data to prove the predication model which the kernel algorithm was mentioned in chapter3. we use the computer simulation and the error analysis to make sure the application and available of the model.