Research On Response Surface Method Optimization Based On Radial Basis Function

It is inefficient and infeasible when using the traditional simulation and optimizationmethods to solve the problem of multi-variable optimization in simulation model, and thedisadvantage is particularly evident in the case of high-dimensional. The response surfacemethod(RSM) based on the experimental design can effectively reduce the number ofsimulation on initial-model in the optimization process, and improve the efficiency ofoptimization design on complex models. Therefore, the RSM has received extensiveattention. Starting on the RBF (Radial Basis Function) interpolation method, this thesiswill research on the RBF response surface method and the RBF-based global optimizationalgorithm.Based on the radial function and using the sample datas as interpolation nodes, theRBF response surface can be easily constructed. There are also some advantages: theinterpolation function is uniquely determined; the algorithm is so simple that it can beeasily implemented by the computer; the performance is outstanding in high-dimensionalnonlinear system. Currently, there are many global optimization methods including thedeterministic approach, the meta heuristics (evolutionary) method, the heuristic directsearch methods, the black-box based response surface optimization methods and so on.This thesis will focus on researching the experimental design based response surfaceglobal optimization method in order to structure a response surface with enough precisethrough less sample datas, and use response surface techniques to reduce thecomputational cost, then combine with rapid response surface reconstruction methods andimproved optimization method to obtain the most optimal solution. Currently a variety ofglobal optimization methods that based on the response surface have the main differenceson three areas including experimental design, construction method of response surface andthe optimization method.when facing to the complex objects whose optimal solution is on the edge ofboundary conditions or other difficult optimizing problems, some of the existing responsesurface-based global optimization algorithm’s performance is not satisfied. Therefore thisthesis introduces an incremental sampling methods based on LHD and a restart strategy. At the same time, proposes an incremental RBF response surface reconstruction methodsto ensure the accuracy and effectively reduce the time consumed by response surfaceupdate, then combined with the CORS optimization method to form a modified globaloptimization algorithm. Finally, the advantages of the improved algorithm are proved withexperiments and the improved global optimization method is applied to an engineeringoptimization example.