Research On Optimal Design Method Based On Adaptive Surrogate Model

With the refinement of simulation experiments,the acquisition of target responses in structural optimization design is becoming more and more time-consuming.Although the use of agent models to assist in optimization design can simplify the calculation,the construction efficiency of agent models and the accuracy of suboptimization problems in agent models need to be further improved.In this thesis,two main perspectives are investigated: 1)improving the efficiency of high precision agent model construction;2)improving the accuracy of the suboptimization problems of the agent model.The main research contents include.(1)From the perspective of improving the efficiency of high-precision agent model construction,an adaptive agent model construction method based on sparsity density and local complexity is proposed,the concept of sparsity density is introduced and its calculation method is improved,the concept of local complexity and the quantification method are proposed,and the points with the largest sparsity density and local complexity are selected to join the sample library to update the agent model and gradually improve the model accuracy.The proposed method is tested by test functions and compared with two high-precision agent model construction methods.The results show that the proposed method can construct an agent model with fewer sample points to meet the accuracy requirements and effectively improve the modeling efficiency.The proposed method is applied to the optimization of the thermal management system of a liquid-cooled cylindrical battery,which improves the performance of the thermal management system and proves the effectiveness of the proposed method in engineering practice.(2)From the perspective of improving the accuracy of the agent model suboptimization problem,the basic genetic algorithm is improved and an agent model suboptimization genetic algorithm is proposed.This algorithm designs a high-quality initial population generation method for the characteristics of the agent model suboptimization function-each existing sample point is the most nearly of the suboptimization function,and the existing sample points are local optimal solution space between them.The roulette wheel method is used in the selection algorithm to ensure the genetic diversity of the genetic algorithm and avoid "premature maturity".The crossover operation is improved for the low computational cost of the suboptimization function,and the local search ability of the genetic algorithm is improved by multiple crossovers.In the mutation operation,an elite retention strategy is adopted to ensure that the good genes can be passed on all the time.(3)The two methods proposed in this paper are applied to the structural optimization of a certain type of turbine disc,and high-precision surrogate models are established for the mass,maximum stress and maximum total deformation of the turbine disc.Firstly,single-objective optimization of the three performance functions is performed,and each optimization objective has obtained a significant optimization plan,but it is found that the optimization of the three performance functions cannot be achieved at the same time.Then the three performance functions were optimized with multiple objectives,which significantly reduced the mass and maximum stress of the turbine disk.