Ranking-based Multi-Objective Particle Swarm Optimization:Research And Application

In the past two decades,Evolutionary Algorithm(EA)has evolved its performance in theory and practical engineering because of its simple structure and the characteristics of joining heuristic ideas easily.On the other hand,with the continues transforming of the natural process,expanding of the project,growth of the demand,various complex optimization problems arised,and the analysis of complex design requirements of the system is getting higher and higher.These problems often contain multiple objectives which called the multi-objective optimization problem(MOOP).However,the traditional optimization techniques such as mathematical programming,gradient descent are difficult to solve.The heuristic-based swarm intelligent optimization method can provide a set of candidate solutions in a single run without the priori information of the problem,which makes the swarm intelligent optimization becomes a natural and effective framework for solving multi-objective optimization problems.Particle Swarm Optimization(PSO)has a good application in the single and multi-objective optimization problems by following the individual's historical optimal and global optimal solution in the search space.However,when solving multi-objective optimization problems,the two optimal solutions are difficult to be determined due to the change of spatial relations.The effect of traditional Pareto-based method will be a sharp decline when face of multi-peak,deception and other complex issues.In addition,the lack of theoretical research in particle swarm optimization hinders the development and the application of practical problems.In view of the above problems,this dissertation puts forward several kinds of ranking frameworks based on the nature of domination,which solves the optimal individual selection in multi-objective particle swarm optimization,obtaining the convergence without losing diversity.On this basis,this dissertation combins with the proposed multi-hydropower owner model to optimize the the cascade of small hydropower optimization problem,which provides an innovative way on cascade joint optimization scheduling.The main work of the dissertation includes:Combining with the ranking strategy,a selection framework of optimal solution for multi-objective particle swarm optimization is proposed.In the past,the selection of the optimal solution in the multi-objective particle swarm algorithm is based on the Pareto optimality and the density crowding distance,which is inefficient and ineffective.In view of this problem,this dissertation combines the sorting strategy with the distribution information of individual in solution space,which can be concise and efficient to obtain the optimal solution in the population,showing the comparative of Pareto-based approach in dealing with multi-objective optimization problems.It provides a new idea for multi-objective particle swarm optimization algorithm to deal with multi-objective problem,especially high-dimensional optimization problem.Besides,the current indicators in evolutionary algorithm mostly from the evolutionary perspective,which lack the evaluation of the ranking method,this paper designs an indicator that considers both the convergence and the efficiency.A method based on grid ranking is proposed,which is applied to multi-objective particle swarm optimization.Most of the previous grid-related work are individual-based in the solution space,which can not guarantee the convergence of the algorithm.In this dissertation,we use the grid to characterize the convergence and distribution at the same time.And a multi-objective particle optimization algorithm based on grid ranking is proposed by mapping the coordinates to sort the dominance in the grid coordinate.A method based on global margin ranking is proposed.The traditional Pareto dominance is based on the comparison of the individuals to obtain the dominant information,which is difficult to obtain the clear information in the whole population.Aiming at this problem,this dissertation r proposes a method based on the global margin ranking,and obtains the domination information of the whole population by using the position information of the individuals in the target space.Based on the optimal solution selection framework,a multi-objective particle optimization algorithm based on global margin ranking is proposed,which uses the position information of the individuals in the objective space to obtain the dominant information in the whole population.Based on the framework of the optimal selection,a multi objective particle swarm optimization algorithm is proposed.By using the stochastic process theory,the convergence characteristics of the PSO in stagnation stage is analyzed.Based on the convergence of the standard particle swarm algorithm,the convergence of the multi-objective particle swarm optimization algorithm and the convergence performance of the proposed ranking framework are analyzed,which provides a theoretical support for the research of the algorithm and lay the theoretical foundation for the practical application.A model for multi-hydropower owner is proposed based on the maximum power beneficial of the whole basin.Taking the Lushui River Basin in Jiangxi Province as an example,the proposed multi-objective particle swarm optimization algorithm is solved.Most of the previous work on scheduling is concerned only with the cascade power generation,ignoring the relationship between multiple power generation entities in the cascade,which makes it difficult for theoretical research to be applied.In view of the lack of linkage mechanism in the basin,the management level differences lead to the waste of water resources,and the instability of power plant.We use the multi-objective evolutionary algorithm to solve these problems.It shows the feasibility and practicality of the algorithm,and provides a new idea for cascade joint optimization scheduling.