Improvement Of Multi-objective Particle Swarm Optimization Algorithm And Its Application In Rolling Schedule Optimization

Particle swarm optimization is a typical intelligent optimization algorithm with the characteristic of simple structure,easy operation and fast convergence.However,particle swarm optimization algorithms still has some shortcomings in convergence and diversity in solving complex multi-objective optimization problems,such as multi-peak problems,discontinuous problems,non-uniform problems,parabolic problems,etc.In view of the above defects,this paper proposes improved multi-objective particle swarm optimization algorithms and applies one to the 1250 mm cold rolling schedule optimization of a steel factory in Tangshan.The main research contents are as follows:(1)Pareto swarm optimization is difficult to converge to the true front and nondominated solution set with uneven distribution when dealing with complex multi-objective optimization problems.This paper proposes a multi-objective particle swarm optimization algorithm based on R2 indicator.This algorithm uses the Sigmoid function to adjust the inertia weight and learning factor of the particle update formula adaptively.In addition,the strategy of fast non-dominated sorting combined with R2 indicator is designed to update and maintain the external archive.The standard test functions ZDT,DTLZ and UF series test functions verify the effectiveness of the algorithm,experimental results show that the algorithm proposed in this paper exhibits good convergence and distribution.(2)This paper proposes a multi-objective particle swarm optimization algorithm based on R2 indicator selection mechanism according to the basic research of a hybrid multiobjective particle swarm optimization algorithm.The contribution value calculated by the improved R2 indicator formula is designed to update and maintain the external archive,and the set are completed without additional convergence or distribution strategies.In order to balance the exploration and exploitation ability of the algorithm in the convergence process,the non-liner Cos function is applied to adjust the inertia weight adaptively.In addition,Gauss mutation strategy is used in avoiding the particle falling into local optimum.The standard test functions ZDT,DTLZ and UF series test functions verify the effectiveness of the algorithm,experimental results indicate that a new multi-objective particle swarm optimization algorithm based on R2 indicator selection mechanism can show good performance.(3)Finally,an improved multi-objective particle swarm optimization algorithm based on R2 indicator is adopted to optimize the rolling schedule.Two objective functions of equal load and prevention of slip are designed according to the purpose of production and characteristics so as to enhance product quality and reduce energy consumption.In order to improve the prediction accuracy of rolling force,the rolling force prediction based on deep learning network is adopted.The simulation results show that the optimized schedule reduces the slip rate and balances the load of each rack.It reduces production energy consumption while ensuring product quality.