Application And Research On Prediction Of Subway Passenger Flow Using Mixed Kernel Support Vector Machine

Urban traffic congestion and energy consumption are becoming more and more serious and urban is confronted with great pressure with the accelerating process of urbanization, the continuous expanding process of population and the gradual popularity of private cars. Urban rail transit with the advantages of high passenger capacity, guaranteed punctuality and environment friendly operation becomes a solution to the problems. Passenger flow volume is the basis for schedules operations of the subway operation management department as well as urban rail transit planning count. Hence, it is of great significance to predict passenger flow volume. In this paper a support vector machine(SVM) regression model based on mixed kernels is applied in the prediction.Firstly, based on the fundamental principles of statistical learning theory, VC dimension theory and structural risk minimization principle are introduced. And SVM regression and the performance of different kernel functions are analyzed. Polynomial kernel function and the polynomial kernel function are linearly combined to construct a hybrid kernel function of which the earning and generalization performance are also analyzed.Secondly, the passenger flow on guangzhou metro line 3 is statistically analyzed to acquire its periodic law and timing characteristics. Furthermore, with a hierarchical cluster analysis of daily checking-in passenger flow, the data sample is divided effectively to provide appropriate data for the prediction model.Finally, the prediction performance of the SVM regression model depends largely on the selection of relative parameters. They are optimized by the particle swarm algorithm rather than the grid test method which has disadvantages of long time-consuming and low efficiency. The infinite folding iteration chaotic mapping is employed to overcome the early-maturing problem of the particle swarm optimization(PSO) algorithm. Meanwhile, the golden section algorithm is employed to improve the optimization speed. Two simulations of testing typical functions can prove the good performance of the modified algorithm which is improved by the combinations of the two methods above. Chaotic particle swarm optimization(CPSO) based on infinite folding iteration and golden section is combined with SVM prediction model based on mixed kernels to construct a combinational prediction model. The checking-in passenger flows at different periods of time are simulated with SVM prediction model based on a single kernel, CPSO-SVM prediction model based on mixed kernels and the combinational prediction model. The result indicates that the combinational prediction model to be constructed in the thesis is prior to the other two.