Research On Surrogate Based Adaptive Sampling And Assisted Evolutionary Algorithms

Surrogate based design and optimization(SBDO)have important applications in solving complex engineering design problems,including Design of experiment determines the reasonable distribution of sample points in the design space,Surrogate modeling methods which establish the approximation between design variables and objective,and surrogate based optimization which searches for the optimum of the problem.In this paper,surrogate based adaptive sampling and infilling strategy are deeply studied,parallel sampling methods are proposed to make full use of the parallel computing resources.To deal with high dimensional expensive problems,combining surrogate models and evolutionary algorithms is an efficient method.The corresponding methods are applied to the design and optimization of underwater vehicles.The main research results are as follows:(1)In order to improve the efficiency of adaptive sampling and make full use of the parallel computing resources,a surrogate based adaptive and parallel sampling algorithm(APS)is proposed.First,the design space is divided based on existing sample points according to Voronoi Diagram method,and each cell contains a sample point.The cross validation error of the sample point in each cell is used as the error criterion,and the size of each cell is used as the sparsity criterion;the mechanism formed by the two criteria determines the target cells,and one new point is selected in the target cell.Parallel sampling mechanism can identify multiple target cells during each iteration,and selects one point in each cell.The selection of each new sample point takes into account the distances to the existing sample point in this cell and previously added points simultaneously by maxmin criterion,which aims at avoiding cluster.Compared with single point sampling method,APS can reduce the number of iterations of sampling process,improve the sampling efficiency.The method is compared with other adaptive sampling methods and the results show that the sampling efficiency is improved.(2)For most surrogate based optimization algorithms,only one sample point is selected by infilling criterion per iteration,which is inefficient when parallel computing resources are available.In this section,several existing infilling strategies in surrogate based optimization algorithms are combined.The performance of these combinations on several test functions is investigated and the most effective parallel infilling stratery is concluded.Then,this parallel infilling combination is applied to the optimization of a location hole on the surface of the underwater vehicle shell.Through the analysis of the optimization results,the general guiding principles for the design of location hole are given.Besides,this parallel infilling combination is also used for the optimization of the hydrofoil of an underwater glider,optimization results show that lifi-to-drag ratio(L/D)is significantly improved.(3)For complex engineering problems with design variables increasing and simulation time-consuming,an evolutionary sampling assisted optimization algorithm(ESAO)is proposed for high dimensional expensive problems.ESAO can deal with the problems with 20 to 50 design variables and the maximum number of function evaluations is 1000.ESAO is consisted of global search and local search.Global search employs all the sample points to build a global surrogate model.The global model prescreens the offspring individuals generated by a differential evolution algorithm and selects the most promising one as the new sample.Local search selects a certain number of best sample points to establish a local surrogate model.Better solution could be found through optimizing the local model.Either search is conducted based on a mechanism and gradually searches for better solutions.The searching mechanism of ESAO is understood through comprehensive analysis.ESAO shows its superiority when compared with other surrogate assisted evolutionary algorithms(SAEAs).The effectiveness of ESAO in engineering application is also verified through the optimization of an airfoil.(4)ESAO is applied to the shape design of a new blended-wing-body underwater glider(BWBUG).The optimization goal is to maximize the lift-to-drag ratio(L/D)at the free-stream condition of ?(28)4,V(28)1m/ s and the constraint is that the volume is not reduced.The problem contains 23 design variables,including 8 plane variables and 15 sectional variables.Establish an automatic calculation framework which can calculate the L/D of a glider through simulation.The mamimum number of function evaluations is 700.Optimized lift increases significantly and optimized drag increases a little at the same time.The L/D increases by 3.64%.The geometry of the optimized glider shows some differences: the size of plane shape increases,especially for the chord length on the wing,which helps increase lift.The thickness of the each section is reduced generally,and the overall thickness of the glider reduces,which contributes to more reasonable pressure distribution.The increase of the drag mainly comes from the increase of surface area.This work provides guidance for the prototype manufacturing of BWBUG.