The Study Of The Swarm Intelligence Optimization Algorithm Based On Trust Region In The Surrogate Models

Swarm intelligence optimization algorithm is an effective method to solve the complex engineering optimization of the high dimension and extreme value problems.The algorithm has strong robustness,extensive application and parallel computing,which can be get effective solution in most cases,is a research hotspot in recent years.However,the group intelligence optimization algorithm is also faced with a problem in dealing with complex problems,that is,in the search for the optimal solution,the need for a large number of complex fitness calculation,time-consuming,affecting the convergence efficiency of the algorithm.In many complex optimization problems,the exact calculation of fitness is very difficult,so we need to find a method of fitness approximation.The surrogate model is a method to calculate the accuracy of the problem by using the approximate estimate of the problem.It can reduce the computational cost and improve the algorithm optimization rate.One of the keys to solve the problem by applying the surrogate model approach is the estimation accuracy of the surrogate model.In this paper,the artificial neural network is used as the surrogate model to estimate the fitness,and an improved trust region method is introduced.The improved trust region method and the group intelligence optimization algorithm are combined to enhance the diversity of the algorithm and improve the estimation of the surrogate model accuracy,so as to improve the optimization efficiency of swarm intelligent optimization algorithm.Particle swarm optimization(PSO)is one of the typical algorithms in the group intelligent optimization algorithms.Based on the particle swarm algorithm algorithm,the simulation experiment is carried out.The main work is as follows:1.A PSO-RBF neural network surrogate model based on trust region is proposed to estimate the fitness value.Firstly,the method of trust region is improved,and the derivative calculation in the process of the trust region searching is replaced by the difference quotient calculation,which avoids the problem of algorithm failure due to the objective function is not differentiable in some practical problems.Secondly,the improved trust region method combined with particle swarm optimization algorithm is used to update the sampling process.And the weights of the RBF neural network are trained by the newly obtained samples to improve the estimation accuracy of the RBF neural network agent model.Finally,based on the PSO algorithm,the simulation results show that the model extends the application range of the algorithm and improves the efficiency of the algorithm.2.An improved PSO-RBF neural network surrogate model is proposed for fitness estimation.In order to improve the estimation accuracy of the radial basis function model,firstly,the connection weight of the radial basis function neural network is trained by the particle swarm optimization algorithm,which avoids the problem that the network is easy to fall into the local optimum when solving the problem.And the optimized network is combined with the improved trust region method to improve the estimation accuracy of the PSO-RBF surrogate model.The simulation results show that the algorithm reduces the number of computations for the fitness,improves the estimation accuracy,further improves the performance of the algorithm,and improves the convergence efficiency of the algorithm.3.A PSO-BP neural network surrogate model based on trust region is proposed to estimate the fitness value.This method uses the BP network with better prediction ability to establish the approximate surrogate model,and uses the particle swarm optimization algorithm to optimize the connection weights and thresholds of the BP neural network to further improve the ability of the network to estimate the fitness value.At the same time,it is combined with the improved trust region method to improve the accuracy of the agent model and reduce the calculation of the fitness.The numerical results show that the algorithm has better global optimization ability.