Research And Application On Approximation Model Based On Support Vector Machine For Structure Optimization Design

Aiming at the time-consuming problem of large-scale structureoptimization, a method (SVM-GA) which combines Support VectorMachine (SVM) and Genetic Algorithm (GA) is proposed. In this method,approximation model is constructed based on SVM to substitute thetime-consuming FEA. GA is applied to optimization. In this thesis, asatellite structure and a ship tank section structure are adopted asexamples to illustrate the effectiveness of the SVM-GA method.Approximation models based on SVM are constructed to simulate themapping relationship between structure design size and response ofstructure. Whereby structures are optimized based on approximationmodel. The main research work is as follows:1. Approximation model methods of optimization design areexpounded. The theory of Support Vector Machine and Support VectorRegression is introduced. SVM-GA method is proposed.2. Satellite structure and its FEM are introduced. Static analysis, modal analysis and frequency response analysis of satellite structure arecalculated via MSC.Nastran. Multi-objective dynamic optimizationmodel of satellite structure is constructed.3. Approximation model of satellite structure based on SVM isconstructed in MATLAB. Based on this approximation model, thesatellite structure is optimized in iSIGHT software. Results ofapproximation model and FEA at optimal solution are compared. Thetime consumption of SVM-GA method is calculated.4. Approximation model of ship tank section structure based onSVM is constructed in MATLAB. Based on this approximation model,the ship tank section structure is optimized in iSIGHT software. As acontrast, the optimization result based on FEA is calculated.Results show that the SVM-GA method can reduce the timeconsumption of optimization and possess high precision. SVM-GAmethod provides an effective solution to the time-consuming problem oflarge-scale structure optimization.