Balance Based Multiobjective Evolutionary Model And Its Application Research

There are a large number of multi-objective optimization problems with multi-standard,multi-constraint and highly nonlinear and conflicting features,in the scientific research and engineering optimization design.The optimal solution of multi-objective optimization problem has non-unique characteristics,and the evolutionary algorithm can obtain a set of solutions in one search.Therefore,evolutionary algorithm is an effectively method for solving multi-objective optimization problems.At present,the analysis and exploration of multi-objective evolutionary algorithms has gradually become a research hotspot.One of the key points of its research is to balance the ability of local exploration and global development,namely double "E",(Exploration and Exploitation).Based on the adjustment mechanism of chemical reaction optimization algorithm exploration and exploitation strategy,this paper mainly study the balance problem between local exploration and global development in multi-objective evolution models which mainly include evolution algorithm operator improvement,learning strategy guidance and multi-directional search for multi-objective evolution model and application.The proposed a balance based multiobjective evolutionary model and its application research applied to solve complex continuous multi-objective optimization problems,multi-objective vehicle routing optimization problems and dynamic multi-objective optimization problems.The main contributions of this work can be summarized as follows:1.Since different operators are suitable for solving different optimization problems,an improved evolutionary operators based multi-objective chemical reaction algorithm is proposed to solve the complex multi-objective optimization problem with variable correlation.In the algorithm,firstly,chemical reaction optimization algorithm is used as a carrier to propose a decomposed multi-objective chemical reaction evolution algorithm.Secondly,in order to enable the basic chemical reaction evolution algorithm to effectively solve the correlation between variables of complex multi-objective optimization problem,an extended chemical reaction is proposed.Finally,the performance of proposed algorithm is compared with eight multi-objective evolutionary algorithms on two sets benchmarks.The experimental results show that the basic multi-objective chemical reaction optimization algorithm has better performance metric on solving the problem of variable non-correlated.However,facing the correlation between variables of complex multi-objective optimization problem,the basic multiobjective chemical reaction algorithm cannot search whole Pareto front.The proposed extended multi-objective chemical reaction optimization algorithm shows better performance metric on solving the variable correlation and variable non-correlated benchmarks.2.Inspired by the particle swarm algorithm learning guidance strategy,a learning-guided hybrid multi-objective chemical reaction algorithm is proposed.In the algorithm,firstly,the self-organizing method is used to divide the evolutionary population into several subgroups according to(m-1)-dimensional piecewise continuous manifold attribute of the Pareto front.Secondly,in order to accelerate the rapid convergence performance of the chemical reaction evolutionary algorithm,a hybrid chemical reaction algorithm with global guiding of particle swarm optimization is proposed.The hybrid algorithm combines the global and local guidance algorithms of the particle swarm optimization algorithm to accelerate the convergence of the chemical reaction optimization algorithm.Furthermore,the local optimality of the hybrid evolutionary algorithm is selected from current group,which enhance the diversity of a molecular/particle on evolution search.Finally,the performance of proposed algorithm is compared with other 23 multi-objective evolutionary algorithms on different a large number of benchmarks.The experimental results show that the proposed algorithm has significant performance compared with the comparison algorithm in terms of convergence,diversity and robustness.3.For the strategy of central point prediction,the whole Pareto optimal set problem of complex dynamic multi-objective optimization cannot be completely predicted.The multi-direction prediction search method is proposed to improve the accuracy of the algorithm prediction.Firstly,the exponential smoothing method is used to predict the next moment Pareto set according to the time series of historical evolutionary population.Secondly,in order to more accurately predict the Pareto set at the next moment,the self-organizing method is used to divide the evolutionary population into several clustering groups.In each group,there has a clustering center,which is computed by center of the group,and then computing direction of center point.Multiple directions guide the evolutionary population more accurately searching whole Pareto front.In addition,in order to maintain the diversity of the evolutionary group,a group of individuals is randomly initialized to maintain the diversity of the predicted population.Finally,the results of proposed algorithm are compared with four different dynamic multiobjective evolutionary algorithms on 12 different dynamic multiobjective optimization problems.The experimental results show that the proposed multi-direction search balance rule can maintain better fast convergence and robustness when dealing with dynamic multi-objective optimization problems compared with the four dynamic multi-objective prediction algorithms.4.A discrete multi-objective chemical reaction algorithm is proposed for multi-objective vehicle routing problem for simultaneous delivery and pickup with time windows.First,for the problem attribute,the solution is encoded in decimal encoding.Secondly,power transformation decomposition method is used to solve actual multi-objective optimization problem which Pareto frontier is unknown.Then decomposition-based multi-objective chemical reaction evolution algorithm is proposed.Finally,the performance proposed multi-objective evolutionary algorithm compares with other compared algorithms on 45 actual benchmarks.The experimental results verify that the multi-objective chemical reaction optimization algorithm based on decomposition has better advantages.