Deep Deformation Prediction Model Of Improved Support Vector Machine

With the rapid development of urbanization, the construction of the multi-storey andhigh-rise buildings and exploitation of a large number of underground spaces, a large number ofdeep foundation pit engineering continues to spring up. Deep foundation pit deformationmonitoring and prediction is an important link of the deep foundation pit design and construction,and is also a hot issue of foundation pit engineering research. To accurately predict thedeformation of the deep foundation pit in the future is the ultimate goal of deep foundation pitdeformation monitoring. In view of the traditional common prediction method contains certainlimitation, based on support vector machine current research situation, the author proposes tosolve effectively support vector machine problems such as small samples, nonlinear, highdimension and local minimum apply in deep foundation pit deformation prediction has beenproposed.The article, firstly, discusses the significance of deep foundation pit deformation prediction,explains deep foundation pit deformation prediction current research comprehensively, analyzesthe deformation of deep foundation pit construction process, proposes predict error minimum todetermine the sample set embedding dimension and time delay, realizes the structure of thesample data, In view of the traditional support vector machine prediction model parametersdifficult to determine problem, the author proposes to use particle swarm optimization algorithm,optimizes related parameters of support vector machine through group random initialization,fitness function set, particle update and the termination conditions set, then gets improvedsupport vector machine forecasting model based on particle swarm optimization.Secondly, the article combined with the deep foundation pit retaining pile body two differentdepth instances the examples survey data, according to the prediction error minimum method towork out training sample set minimum embedding dimension and time delay, the authorreconstructs coordinate deformation data sequences to coordinate delay in phase space, usesphase space domain phase point, constructs space structure to get learning samples, Then, theauthor combines with Microsoft Visual c++6.0compiler on Matlab7.14platform, uses libsvmtoolbox to extend programming to realize training and prediction of traditional support vectormachine model and improved support vector machine model.Finally, according to the organizational Matlab program, the article makes prediction resultthe support vector machine forecasting model and traditional support vector machine model andElman dynamic neural network model, using the mean square error, squares sum error, theaverage relative error to evaluate the prediction effect, then it gets the mean square error, squaressum error, the average relative error of improve support vector machine prediction model are respectively0.0155and0.0164,0.0155and0.0164,1.2511percent and4.2205percent.Theexperimental results show that based on improved support vector machine forecasting modelprediction mean square error sum of squares of the average relative error is superior to traditionalsupport vector machine model and the Elman network model, through the particle swamoptimization support for optimal support relate parameter of vector machine forecasting model, itcan be better improved support vector machine forecasting model, and has better fitting effect,generalization capability, stable performance, high prediction accuracy, it proves that based onimproved support vector machine forecasting model can better reflect the dynamic nonlinearcharacteristics of deep foundation pit system, it has certain superiority and the engineeringapplication promotion value.