Research On Swarm Intelligent Optimization Methods And Their Applications In Chemistry And Chemical Enginnering

In recent years,with global economic competition growing,increasingly stringent environmental regulations,the continuous improvement of the technical and economic indicators,as well as the non-renewable resources dwindling,etc.,the architecture of the chemical industry has undergone tremendous changes.The concept and definition of process systems engineering has been increasingly broadening,and process systems engineering is concerned with the improvement of decision making processes for the creation and operation of the chemical supply chain.Ranging from product development to supply chain management,optimization techniques can be implemented in any scale of chemical process. Moreover,lots of new ideas,such as product engineering,molecular engineering,green process systems integration,multi-scales modeling,etc.,have been introduced into traditional chemical engineering disciplines;therefore the constructed optimization models will be more complex,which-may include nonlinearity,dynamic,combinatorial,multi-objective and uncertainty,the conventional optimization methods often failed,so the demand for efficient intelligent optimization methods is becoming more urgent.Swarm intelligent optimization method is an emerging area of research,which provides an effective tool for solving complex optimization problems,and it has attracted extensively attentions from different research areas.Ant colony algorithm and particle swarm optimization are two typical kinds of swarm intelligent optimization methods,and they have achieved great successful applications in combinatorial optimization field and continuous optimization field respectively.However,their applications in chemistry and chemical engineering are relative less,and both of them lack effective strategies for dealing with such optimization problems that include constraints,dynamic and multi-objective.Hence,the main research work of this dissertation includes two-fold contents,on the one hand,extending and improving swarm intelligent optimization methods make them suitable for handling different types of optimization methods;on the other hand,expanding their applications in chemistry and chemical engineering.Aim at the shortage of traditional ant colony algorithm,which is just suitable for solving discrete optimization problems,three kinds of continuous strategies are proposed in this dissertation,which can be suitable to continuous optimization problems.Moreover,the proposed continuous ant colony algorithms are further extended to handle such optimization problems with multi-objective,dynamic,and constraints.The proposed methods are applied to practical optimization problems stemming from chemistry and chemical engineering,the obtained results illustrate that proposed ant colony algorithms have well adaptability and global optimization performance,and it is convinced that they have greater potential applications in process systems optimization area.Besides,two kinds of particle swarm optimization algorithms are modified and proposed to deal with mixed-integer nonlinear problems and dynamic multi-objective optimization problems respectively,and their effectiveness are illustrated by applying to process synthesis cases and dynamic multi-objective optimization of biochemical reactor.The main contributions of this work are summarized as follows.[1]Aim at the shortage of traditional ant colony algorithm,which is just suitable for solving discrete optimization problems,three kinds of continuous strategies,such as instance-based,recruitment-based and model-based,are proposed.The instance-based continuous ant colony algorithm,called hybrid ant colony system(HACS),adopts genetic operation for global exploration and powell method for local exploitation;HACS has been successfully applied to estimate the kinetic parameters of 2-chloropheo oxidation in supercritical water.The recruitment-based ant colony optimization called MG-CACO adopts group and mass recruitment for achieving positive feedback and the phenomenon of anorexia for achieving negative feedback;MG-CACO has been successfully applied to the single objective operation optimization of RBF-MCSR model for the equipment of xylene isomerization.Moreover,MG-CACO has been extended and modified to handle constraint optimization and multi-objective optimization problems.For handling constraints,firstly three heuristic rules for evaluating the quality of food source are embedded into the MG-CACO algorithm framework;secondly the mass recruitment operation and pheromone update operation,are modified;finally,a novel method called MG-CCACO is proposed to constrained continuous optimization problems.MG-CCACO has been successfully applied to process constrained optimization of butane alkylation.For handling multi-objective,firstly, through comprehensive consideration of dominated degree and dispersion degree,the heuristic rules are proposed to evaluate the quality of food source.Secondly,an extemal archive technique is set up for reserving non-inferior solution set,and an appropriate update operation of archive is adopted to uniformly approximate the Pareto optimal solution set step-by-step.Finally,a novel algorithm called MG-MOCACO is proposed to multi-objective optimization problems.MG-MOCACO has been successfully applied to the multi-objective operation optimization RBF-MCSR model for the equipment of xylene isomerization.The model-based ant colony algorithm called IM-MOACA firstly adopts a mixture of normal distribution to describe the pheromone distribution.Secondly,both the concepts of Pareto dominate and concentration of immune system are adopted to evaluate the solution quality. Thirdly,pheromone update is based on both the external archive and the current population. Finally,a crowding distance based external archive update strategy is adopted to uniformly approximate the Pareto optimal solution set step-by-step.IM-MOACA has been successfully applied to two dynamic multi-objective optimization problems of batch reactor.[2]The rule based knowledge is easy-to-understand and is in clear correlation with the professional knowledge.However,the classification rule extraction is a difficult task and it often needs a discretization as a pre-processing step for continuous attributes.In this work, two novel approaches are proposed to rule extraction.(1)The problem of classification rule extraction is considered as a constrained continuous optimization problem and MG-CACO, which can automatically extract the classification rules without discretization as a preprocessing step and can capture the interdependencies between attributes,is proposed to extract classification rules.Moreover,for improving the prediction performance,an ensemble strategy is adopted,and the ensemble classifier system called MG-CACO-ECS is built. MG-CACO-ECS has been successfully applied to the producing area discrimination of olive oil.(2)Based on constructed distinction table,continuous attributes discretization and reduction are synthesized into a bi-objective optimization problem,and a compromised-based ant colony algorithm called CACA is proposed to solve this bi-objective optimization problem. CACA calculates heuristic information dynamically during solution construction;both redundant column removing operation and local search operation are adopted to improve the convergence speed;two kinds of pheromone update operations are implemented with probability,which can balance the trade-off between global optimization ability and convergence speed.CACA has been successfully applied two toxicity mechanism classification problems:the classification of three narcosis mechanisms of aquatic toxicity for 194 organic compounds and the classification of four action modes of 221 phenols.[3]Aim at the shortage of traditional particle swarm optimization,which is unsuitable for handling constraint,discrete variable and multi-objective,two kinds of modified strategies are proposed to deal with MINLP problems and dynamic multi-objective optimization problems.(1)A hybrid particle swarm optimization called HPSO is proposed to solve MINLP problems.HPSO introduces the concept of Pareto dominate to evaluate the solution quality,a random rounding method based on distance function is introduced to dealing with integer variables,and stochastic mutation based solution repair strategy was embedded for increasing the convergence speed;moreover,both multi-swarms strategy and velocity update strategy, utilizing all particles' local best information,were adopted to enhance the population diversity. HPSO has been successfully applied to six process synthesis case studies.(2)A multi-objective particle swarm optimization called MOPSO is proposed to dynamic multi-objective problems.Firstly,the concept of Pareto dominance is used to evaluate the fitness of particles,and two kinds of operation for determining the local and global optimal point respectively are designed.Secondly,setup an external archive technique,and through calculation of dispersion degree,an appropriate update strategy is adopted to uniform approximate the Pareto optimal solution set step-by-step.MOPOS has been successfully applied to two dynamic multi-objective problems of induced foreign protein fed-batch production process...