Research Of Intelligent Algorithm Application In Chemical Dynamic Optimization

Dynamic optimization problems(DOPs)are common in the chemical production process.Effectively solving DOPs can improve the production efficiency as well as reduce the raw material consumption.However,the optimization variables of DOPs are continuous variables,and also have complex characteristics such as multi-peaks and multi-variables,which make the DOPs difficult to solve.This thesis conducts research on dynamic optimization methods based on the intelligent optimization algorithms(IOAs),under the control vector parameterization(CVP)framework.The main research content is as follows:(1)Conducting a comparison study of eight typical IOAs in solving chemical DOPs.Previous studies mainly focused on improving the performance of existing IOAs,while ignoring the performance comparison among these algorithms.Therefore,we carry out a research to compare eight typical IOAs in solving the chemical DOPs.Through solving four DOPs,the performances of different IOAs are comprehensively analyzed from the three aspects,i.e.,statistical accuracy,Friedman ranking,and convergence speed.By analyzing the results,it is concluded that the differential algorithm(DE)has better performance than the other seven algorithms when solving these DOPs.(2)Developing a hybrid gradient adaptive difference evolution algorithm for solving chemical DOPs.In order to improve the performance of the IOAs in solving DOPs,by combining the adaptive DE algorithm and the gradient search algorithm,a hybrid gradient adaptive difference evolution(HGADE)is presented to solve DOPs.HGADE is divided into two parts.First,the adaptive DE algorithm is used for global optimization,and then gradient search is used to improve the search accuracy.Under the CVP framework,HGADE is used to solve four DOPs.The results show that HGADE has advantages in solution accuracy and other metrics.(3)Presenting a sigmoid basis function based CVP approach for solving chemical DOPs.In the past,CVP framework mainly uses constant or linear basis function to appropriate the control curve,which has the disadvantage such as discontinuity and non-differentiability.Taking advantage of the continuous differentiability of the sigmoid function,three types of sigmoid functions are stitched together as an equivalent control curve.Using the sigmoid function in CVP to solve six DOPs,the control curve has continuous and differentiable advantages compared with the solution of constant or linear basis function.Each control curve can be adaptively approached constant or linear equivalent curves have more advantages in solving multivariable DOPs.