Implementation And Application Of Optimization Algorithm For Black Box Model Problem

In chemical industry,optimization techniques are widely used to reduce costs and enhance efficiency.When optimized an industrial process,the first step was always to build a complete optimization model of the industrial problem,and then use the model information for efficient optimization calculations.However,the problem of opaque model is often encountered in actual operation,namely,it can produce input and output information but lacks an open model description.This paper focuses on the problems of black-box model to study and develop an efficient optimization algorithm.The main content of this paper is as follows:1.Implemented optimization algorithms for the black box models.First,according to the characteristics of the black box optimization model,this paper analyzed different methods and chose the model-based derivative-free optimization method to implement the algorithm.The sampling and interpolation method based on the model-based derivative-free optimization theory provides a local approximation model of the original problem.Then combine this with the traditional trust-region optimization method,regarding it as a sub-problem within the trust region,and finally implemented the solution of black box models with the above mentioned trust-region optimization framework.2.Improved this paper,s optimization algorithm with the reference from the NEWUOA framework,then analyzed and compared the performance of the optimization algorithms.The NEWUOA algorithm proposes a method for constructing a quadratic model using less than(n+1)(n+2)/2 interpolation points,which has achieved good performance in numerical experiments.This paper use this idea to make modifications in this paper's algorithm,which reduce the number of function evaluations during optimization,and enhance the generality of the algorithm.Finally,the test results of of five different algorithms were compared and analyzed.3.The NEWUOA-based derivative-free trust region optimization algorithm was applied to solve the industrial black box problems.First,this paper used the optimization algorithm to automatically tuning the PID parameters of a single-sink control system,and compared it with the traditional tuning.Then it was applied to the optimization model constrained by molecular weight distribution specification for polymerization process.Compared with the previous three algorithms,the experimental results show that the NEWUOA-based derivative-free trust region optimization algorithm can achieve the highest conversion rate,and have the least computational burden.