The Research And Application Of Improved Particle Swarm Optimization Algorithm In The Multiobjective Optimization Problems

Multi-objective problems with constraints exist widely in real life. Among them,grinding and classification process is a key part of the beneficiation process, whichdirectly affects production quality and economic benefits of the entire plant, but isalso a complex, non-linear multi-objective problem. So, it’s of very significance thatan effective multi-objective optimization algorithm is developed to optimize thegrinding and classification process.It is difficult to solve optimization problems with the discontinuous, nonlinearand other complex characteristics by traditional optimization technology.Evolutionary algorithm provides a new way to solve multi-objective optimizationproblems and multi-objective evolutionary algorithm has been one of the focus areasin nearly20years. With the development of evolutionary algorithms, many classicalgorithms have been proposed. Among them, particle swarm optimization (PSO) hasalways been one of the hot topics in recent year owing to its advantages, such assimple concept and easy combination, high efficiency, etc. And PSO has beenexpanded for solving multi-objective optimization problems.As dynamic multi-swarm particle swarm optimization algorithm (DMSPSO)was proposed, the research about PSO upsurge again. In DMSPSO, the swarms areused to strengthen the local search ability and dynamically reconfigured to increasethe diversity of population. Special updating mechanism for individual and localoptimal solution can speed up convergence speed and improve effectively thediversity of population. It has been applied successfully in various areas and is anefficient multi-objective optimization algorithm. Aiming at the shortcoming of fallingeasily into local optimum, on the basis of DMSPSO, some disturbance is increased toexternal set in order to further strengthen the ability to jump out of local optimum.With comparison to other classic algorithms on unconstrained standard test functions,the simulation results showed that the modified algorithm is effective.To solve the optimization problems with constraints, based on the improvedDMSPSO, double-file strategy and improved feasibility-based rules are added. The feasible solutions and infeasible ones are kept separate to reduce the numbers ofcomparisons between the particles, thus the infeasible solutions which have smallvalues in violation of the constraints will be kept to improve the ability of using theinfeasible solution. Algorithm is verified efficient through the simulation in standardtest functions.Finally the proposed algorithm will be applied to the real constraint optimizationproblem-grinding process. Multi-objective optimization models were cited from aliterature which used the similarity criterion and the decision-making method basedon TOPSIS were applied to find the optimal value for optimal control of grindingprocess. The result is helpful to guide the actual industrial process in theory in orderto improve production efficiency and reduce energy consumption.