Method and device for constrained optimization of proxy model based on multi-point addition

Abstract

The present application relates to a method, a device, a computer device and a storage medium for constrained optimization of an agent model based on multi-point addition. The method includes: setting parameters through the time-consuming function information of the object to be optimized, performing initial sampling, evaluating the function value and constraint value of the sample point, and establishing a Gaussian process proxy model of constraint and target based on the existing sample library. When there is no feasible solution that satisfies the constraints, the optimization algorithm is used to optimize the feasible probability function to obtain the optimal solution and add it to the sample library until there is a feasible solution that meets the constraints in the sample library. Construct a multi-objective optimization sub-problem, and take the expected improved local mining and global exploration characteristics penalized by the feasible probability function as two objective items; solve the multi-objective optimal solution set including multiple candidate points, when not When the convergence criterion is met, multiple optimal points are selected from the candidate points as new samples, and the Gaussian process proxy model of the objective function is updated until the convergence criterion is met, and the optimization is completed.

Description

Technical field[0001]The present application relates to the field of engineering design, in particular to a multi-point dotted agent model constraint optimization method, device, computer device, and storage medium.Background technique[0002]In engineering design, it is often necessary to rely on time-consuming numerical simulation. Optimized design issues based on these simulations are often referred to as time consuming optimization problems, such as airfoil design, aircraft shape design, beam structure design, etc. These problems are usually highly non-linear, and the traditional gradient optimization algorithm is easily caught in a local optimal solution; and modern heuristic algorithms typically require a large number of functional assessments, solving such problems is very low. The optimization method based on the agency model is to solve the most effective way to solve such engineering design problems.[0003]Agency model optimization method refers to an approximate mathematical...

AI Technical Summary

Problems solved by technology

CEI can only obtain one update point in one iteration, and cannot use the parallel computing environment to accelerate the optimization convergence process, and it is also difficult to explore unknown areas and tap the local optimal solution in the design space at the same time

The multi-point addition criterion EIvsPF (Expected Improvement versus Probability of Feasibility) uses EI and PoF as two objectives to construct a multi-objective optimization sub-problem, and then selects multiple points from the frontier as update points, but this criterion will cause partial update sample point aggregation The phenomenon of being near the global optimal solution, if the global optimal solution is not located in the feasible region, it will affect the optimization convergence efficiency

[0005] Therefore, the prior art has the problems of low optimization efficiency and large error

Images