Research On MOEA/D Algorithm Based On RBF Surrogate Model

Practical engineering problems often involve multi-objective optimization.The input and output of these optimization problems often need to be calculated by time-consuming simulation experiments,which leads to the failure of conventional optimization methods.The multi-objective evolutionary algorithm based on surrogate model(SMOEA)has become a research hotspot because it can greatly reduce the time cost.However,it still has shortcomings such as low solution precision and poor Pareto solution set performance in large-scale and high-dimensional design space.Based on this,this thesis conducts an indepth study on the approximate model-assisted multi-objective evolutionary algorithm.Based on the existing research,the performance of SMOEA is improved and the time cost of the algorithm is reduced.The main research contents of this thesis are divided into the following three aspects:Firstly,in order to improve the accuracy of the approximate model in the optimal solution set region and reduce the time cost of optimization,an adaptive RBF-based MOEA/D algorithm is proposed.The algorithm adopts the Latin hypercube experimental design method,and makes the initial selection of the entire design space and establishes the initial model.The decomposition strategy is used to decompose the multi-objective problem into a single-object sub-problem.The MOEA/D algorithm is used to solve the optimization information of the current model.The sample points with the best sub-problems are added to the sample library,and the approximate model is rebuilt until the algorithm converges and the Pareto solution set is output.At the same time,the uniform decomposition method is introduced in the MOEA/D algorithm,and the evolution strategy based on the population crowding distance is proposed to improve the performance of the algorithm.Secondly,in order to improve the accuracy of the approximate model in highdimensional space,a variable fidelity RBF modeling method based on linear scale and RBF model correction is proposed.Firstly,a low-fidelity surrogate model is established in the design space by low-reliability response values of a large number of sample points,and a high-reliability response value of a small number of sample points is used to establish a linear mapping relationship between high-and low-fidelity models.The RBF model is corrected to improve the accuracy,and the low-fidelity surrogate model is mapped to the high-fidelity surrogate model.The adaptive modeling method and the MOEA/D algorithm are used to solve the Pareto optimal solution set in the high-dimensional space.Finally,this thesis introduces the impact of vehicle side collisions and national policy requirements,based on the EREC95 regulations and EEVC design,establishes a multiobjective optimization problem.The target is to minimize the side impact performance of the vehicle and the total mass of the main components.The design variables are 11 parameters such as the internal thickness of the B-pillar,the thickness of the B-pillar reinforcement,the thickness of the inside of the floor,and the thickness of the beam.At the same time,the detailed steps of solving the problem by VFRBF-MOEA/D algorithm are given.Finally,the effectiveness and feasibility of the algorithm are verified by comparing the optimization design with the initial design.