Dynamic Evolutionary Modeling And Rbust Optimization Of Decision-making Parameters For Complex Industrial Process

Complex industrial process has the multiple characteristics，which gather the multivariate,strong coupling，uncertainty and big fluctuation of performance index in integral whole.Therefore, they will lead to the enterprise facing serious constraints and challenges onresources, energy and the environment. However, modeling and optimization of complexindustrial processes are one of the keys to improve performance index and conserve energy.The purpose of this study is to model the dynamic evolution modeling of complex systemwhich can adapt to changes in the production process and establish a method of robustoptimization for decision-making parameters using data mining theory. Then, the study willmainly expand as following three aspects taking the case of aluminum electrolysis productionprocess.Firstly, select decision variables of aluminum electrolytic process. A novel method basedon Kernel Partial Least Squares (KPLS), false nearest neighbor method (KPLS-FNN) andrandomization method is proposed to select the most suitable decision variables of industrialprocess. In the proposed approach, the KPLS can be employed to map the original space bynuclear transforma to Partial Least-Squares space; the similarity measure and significancelevel can be calculated respectively by FNN and randomization method, so the variable whichis not significant in the statistical sense will be eliminated.Secondly, establish the dynamic evolutionary mdeling of technical energy consumptionin aluminium electrolysis process. A precise process modeling is the prerequisite ofoptimization. An novel Unscented Kalman Filter Neural Network combined strong trackingflter and square root filter(STR-UKFNN) is proposed in the study. The STR-UKFNN is usedto establish the dynamic evolutionary modeling for technical energy consumption. In theproposed approach, the state covariance matrix of UKFNN algorithm is replaced by its squareroot to participate in recursive operations; meanwhile, the filter gain matrice of UKFNN isadjusted by introducing the time-varying fading factor and the diminishing factor.So, theconstringency speed of modeling and tracking ability for mutation status can be both improved.Thirdly, optimize the decision-making parameters of aluminum electrolytic process. Thedesign and implementation of optimal parameters can ben effected negativly by theuncertainty of the interference from aluminum electrolysis process. A novel method based ondynamic modeling and robust optimization is proposed to design the optimal program. On thebasis of the accurate energy consumption modeling(STR-UKFNN), an improving strengthPareto evolutionary algorithm is used to design optimal Parameters. So, the system can beguaranted to have robust and optimal output parameters which will help reduce energyconsumption.In summary, the study suggests three methods:(1) first one based on KPLS-FNN andrandomization method to select the most suitable technical parameters;(2)second on using theSTR-UKFNN to establish dynamic evolutionary mdeling;(3)third one base on dynamicmodeling and robust optimization to design the optimal program.The research on the modelingand optimization of complex industrial process provides an effective way to improveperformance indicators and achieve energy saving and emission reduction for industrialprocess.