Research On Sequential Optimization Methods Based On Adaptive Accelerated GEI Criterion

With the increasingly high demand on mechanical product performance and growingly high complexity in the mechanical product design optimization process,the sequential optimization method based on the surrogate model becomes more and more widely used.The surrogate model,which replaces the problem of “black box”,can reduce calculation and time costs in the design optimization process.Kriging model,which can know the standard deviation of the fitting model,has unique advantages among numerous surrogate models.It is crucial to determine the next sample point in the design optimization process based on Kriging model.Expected improvement(EI)criteria is widely applied in determining the next sample point,because it balances both global search and local search.But classical EI criteria may produce a smaller error of estimation,which will make the optimization process skew near the current optimum point to search.Thus scholars have proposed a more adaptable generalized expectation improvement criteria(GEI)criteria.Based on sequence optimization of GEI criteria and multipoint sampling problems,this dissertation proposes adaptive speed generalized expected improvement(AGEI)criteria,and applies AGEI to multi-point sampling,applies kriging believer algorithm based on generalized expected improvement criteria to solve the engineering example of power flow optimization of all-direction propeller.Firstly,to solve the problem that current GEI criteria can't accurately determine g value,we propose adaptive speed generalized expected improvement criteria,which can improve the efficiency of solving sequential optimization by taking a large number of random points,calculating EI value of each random point,counting qualified random point number and determining g value according to the model fitting situation.Secondly,scholars focus on the sampling method and reducing the mathematical calculation in current sequential optimization multi-point sampling.This dissertation applies AGEI to multi-point sampling,which can improve solving efficiency of kriging believer(KB)algorithm by combining AGEI with multi-point sampling algorithm-KB algorithm.The high efficiency of AGEI and advantages of parallel computing of KB algorithm multi-point sampling are made a good combination.Finally,by constructing the finite element model of power shafting,establishing the solution strategy of power flow and conducting the finite element analysis of bearing temperature field,this paper successfully establishes a mathematical model using bearing preload as the variable and aiming to solve the minimum power flow as the goal.Kriging believer algorithm based on generalized expected improvement criteria(AGEIKB)is applied to the engineering example of power flow optimization of all-direction propeller,which greatly reduces the power flow into the engine.The feasibility and high efficiency of the proposed strategy are well verified.