Study On The Slope Displacement Prediction By Particle Swarm Optimization

Slope is the maximum and complexest structure when struct the road and other infrastructure,the stability is closely related to the safety of project construction and operation.There are two ways to judge the stability of a slope at the moment,one is from the numerical simulation by slope body and other external factors,another is judge it by forecasting slope displacement over a period of time in the future.In the second method,as the slope displacement is affected by the coupling of internal and external factors,usually show the feature of nonlinear and revulsion,it can't be directly predicted by linear regression method.In view of the feature in slope displacement,in this paper,we extract the revulsion by empirical mode decomposition(EMD),and forecast the slope displacement by support vector machine(SVM)which parameters are acquired by particle swarm difference algorithm(PSO-DV).In order to improve the forecasting accuracy,this article study on the random term extraction,kernel function selection and parameter value scope.Ahead of the study on the problems in process of extraction random items by EMD,analyses were conducted on theoretical basis and calculation process of decomposition method.Combined endpoint value judgment with mirror continuation method solve the extremum continuaton endpoint;in allusion to the problem of less or no extreme value point on slope displacement,propose the method of decomposition stage displacement;study on the reasonable value about decompo-sition cycle by experimental.Compared the trend term with the regular of slope displacement which is under the action of gravity,conclusion show that we can obtion a good decomposition effect during decomposition cycle values for 15 or20.For the process of slope displacement prediction which is based on SVM,analysis the theoretical foundation and calculation process firstly.With the feature analysis of EMD and polynomial kernel function and guassian kernel function,put forward to the method that using polynomial kernel function forecast trend item and applying guassian kernel function forecast random item.As for parameters value problem which included in the displacement prediction of SVM,applying the optimization results of particle swarm difference algorithm.For the value or value range of various parameters,come up with the method that search it by the rule of mean absolute error.As a result: the training cycle of trend item and random item 1,2 is 7,5,6 when there are two random i t-ems,the training cycle of trend item and random item 1,2,3 is 7,5,5,6 when th e-re are three random items;For the scope of SVM parameter,oscillation of slo pedisplacement are shown in table 4.2,growth of slope displacement are shown in table 4.3.At last,we predict the slope displacement with the research result of second chapter to fourth chapter,the distribution regularity of prediction error is analyzed by absolute error and relative error,and come to the conclusion that the proposed method and parameter value range have a high accuracy in oscillation and growth slope displacemengt prediction.