Large-Scale Optimization Algorithm With Its Applications In Supply Chain Network

In the big data era,we always need to deal with massive data,and the scale of the optimization problems is gradually expanding.Therefore,large-scale optimization problems emerge.In large-scale optimization problems,the search space of solutions is larger and more complex,which makes it hard for the traditional optimization algorithms to get satisfactory solutions in the limited time.In order to deal with these problems,this paper designs the large-scale optimization algorithms from two aspects: co-evolution based on the idea of "divide and rule" and overall evolution to enhance the diversity of population.In addition to the theoretical research,this paper also takes the large-scale supply chain network design with uncertainties and large-scale multi-objective supply chain configuration as the objects,and studies the application of large-scale optimization algorithms.The main innovations of this paper are shown as follows:(1)A graph-based re-decomposition method is proposed to improve the decomposition efficiency of large-scale optimization problems.The existing decomposition methods cannot effectively decompose large-scale overlapping problems,and their grouping efficiency depends on the accuracy of interactions among variables(Ia V),which limits the optimization efficiency of cooperative coevolutionary algorithms.In this paper,by dealing with the computational error of Ia V,a recursive decomposition method and a grouping adjustment strategy are designed to divide the variables of the problem into multiple groups with suitable sizes,so as to enhance the fault tolerance and grouping efficiency of the existing decomposition methods.In addition,to verify the grouping efficiency of this method,an overlapping problem generator with the complex Ia V information and two new metrics are designed.(2)An adaptive population control strategy is designed to increase the population diversity and to improve the efficiency of the traditional differential evolution(DE)for large-scale optimization problems.In the selection operation of DE,only if the newly generated solution performs better than the original solution,it will be added to the new population;otherwise,it will be eliminated directly,which will waste computing resources and is not helpful for the algorithm to find the optimal solution of large-scale problems within the limited computing resources.Therefore,this paper designs a population increasing strategy for increasing the probability of new generated solutions to survive in the new generation and to make full use of the evolution of individuals,so as to increase the population diversity.Meanwhile,in order to avoid the population expanding gradually,this paper designs a population decreasing strategy to eliminate the solutions with poor qualities and no improvements for a long time and to assign the limited computing resources to other promising solutions,which helps to speed up the convergence of the algorithm.(3)A cooperative co-evolutionary particle swarm optimization based on function independent decomposition and two repair operators are designed to effectively solve the large-scale supply chain network design problem with uncertainties and complex constraints.Firstly,Monte Carlo method is used to simulate the uncertain factors and to evaluate solutions.Secondly,according to the characteristics of the problem,this paper proposes a cooperative co-evolutionary particle swarm optimization algorithm based on function independent decomposition to decompose the large-scale problem,and uses several populations to solve sub-problems separately,which helps to reduce the search space of each population and the solving difficulty of the problem.In addition,for increasing the population diversity,two repair operators are designed to improve the infeasible solutions and to acquire more feasible solutions.(4)An Efficient Local Search based algorithm with rank(ELSrank)is designed to effectively solve the large-scale multi-objective supply chain configuration problem.Different from the traditional swarm intelligence algorithms for solving multi-objective optimization problems,the proposed local search algorithm is not based on the population evolution.Firstly,this paper designs the calculation method of supply chain members' configuration ranking,and designs a local search algorithm to explore the common neighborhood between two solutions and obtain better solutions in the neighborhood.Secondly,this paper obtains the optimal solutions of the corresponding single-objective optimization problems by the greedy strategy,and based on the greedy solutions,uses the local search to find the optimal solution set and improve it continuously.The local search can obtain the optimal solution set efficiently and accelerate the convergence speed.(5)A multiple-population ant colony system is designed to effectively solve the large-scale multi-objective supply chain configuration problem.Based on the multiple-population multi-objective framework,two ant colonies are used to solve the two optimization objectives of the problem.According to the characteristics of the problem,this paper obtains the priority of supply chain nodes,and designs a greedy heuristic strategy to construct better solutions.Secondly,a local and a global pheromone updating methods are designed based on multiple populations to make the evolution information effectively feedback to the pheromone updating.In addition,this paper uses ELSrank to improve the current optimal solutions,so as to quickly find more promising solutions.To sum up,this paper studies large-scale optimization algorithms from two aspects of algorithm design and the application supply chain network.In the aspect of algorithm design,this paper firstly proposes a graph-based decomposition method to effectively decompose large-scale problems,so as to improve the optimization efficiency of coevolutionary algorithm.Secondly,a population control strategy is proposed to increase the population diversity of DE algorithm.In the other aspect,this paper designs a cooperative particle swarm optimization based on the function independent decomposition to solve the large-scale supply chain network design problem,and uses repair operators to improve the infeasible solutions and enhance the feasibility of solutions.In addition,a local search algorithm and a multiple-population ant system algorithm are designed to effectively solve the large-scale multi-objective supply chain configuration problem from different perspectives.