Research On Improvement Of Multi-objective Evolutionary Algorithm And Its Application In Rolling Load Distribution

Some engineering problems require optimization of multiple conflicting objectives simultaneously,which are called multi-objective optimization problems.These problems are almostly non-linear,non-differentiable problems,which are hard to optimize by traditional optimization methods.Evolutionary algorithms perform better than traditional algorithms when dealing with multi-objective optimization problems with nonlinear,high dimension and complex relationship.However,multi-objective evolutionary algorithms still show some defects when dealing with problems with complex Pareto front or many-objective.Aiming to solve the disadvantages of existing multi-objective evolutionary algorithm,different kinds of multi-objective evolutionary algorithms have been put forward to optimize different types of optimization problems,and a load distribution system is designed for rolling schedule optimization on aluminum hot tandem rolling in Henan "1+4" aluminium strip line.The main contents of this thesis are summarized as follows:(1)Traditional multi-objective particle swarm optimiaztion algorithms focus on evolutioanry operator and selection mechanism,but the improvement of isolated design of evolutionary operator and selection mechanism are limited.To improve the adaptability of particle swarm optimization algorithm for multi-objective optimization problems with different types,multi-obj ective particle swarm optimization algorithm based on combination of decomposition and dominance is put forward.Different selection mechanisms are used to select global best particles and personal best particles,and the relationship between individuals in solution set is integrated.Based on the simulation of ZDT and DTLZ test problems,the proposed algorithm achieves a good adaptability on problem with different type of Pareto fronts.(2)Aiming to solving the multi-objective optimization problems with complicated linkage variables,a multi-objective evolutionary algorithm based on environmental and history information(MOEA-EHI)is proposed.The history information of individuals with good performance are used as evidence to generate offsprings,which improves the generating probabilitiy of elite individuals;the environmental information of reference front is used as the evidence of individual selection,which improves the survival probabilitiy of elite individuals.Based on the combination usage on environmental and history information,searching ability on decision space and divisity preservation ability have been improved.Based on the simulation of UF test problems with complicated linkage variable,algorithm achieves a good convergence on both decision space and objective space.(3)Foucing on solving the failure of Pareto dominated relationship on many-objective problem,a many-objective evolutionary algorithm based on adaptive reference vector is proposed.The algorithm adopts angle cosines of the nearest individual as the value of diversity penalty,and uses angle cosines between individual and reference vector and the location distance between individual and origin as the basis of selection mechanism.Evolutionary population and elite population are selected based on the different selection mechanisms,and information is exchanged in evolutionary process to maintain the diversity of the population.Simulation has been conducted on DTLZ test problem with 5,10,15 objetives,and results show that the proposed algorithm achieves a good convergence.(4)The problem of load distribution on aluminum hot tanden rolling is a typical multi-objective problem.Rolling force is an important parameter of load distribution.To improve forecast precision of rolling force,an artifical neural network based on data classification has been employed.Three different objective functions have been desgined according to process requirement and optimized by multi-objective evolutionary algorithm to improve the production efficiency and quality.The simulation results show that the optimized load distribution scheme reduces the slippage problem of finishing mill group,balances the load distribution,and reduces the total energy consumption of rolling.