Research On Quantum Behaved Particle Swarm Optimization Algorithm And Its Applications For Single-objective And Multi-objective Optimization Problems

Swarm intelligence optimization algorithm is a stochastic optimization algorithm by simulating biological groups foraging,migration and other social behavior.The swarm intelligence optimization algorithm adopts a heuristic search technique for optimization.It does not need the gradient information of the objective function.As one of the representative for swarm intelligence,particle swarm optimization(PSO)algorithm has been studied and applied widely.The PSO algorithm has simple structure,few control parameters and fast convergence properties.However,PSO algorithm is easy to fall into local optimum point and appear premature convergence at the later iteration.The quantum-behaved particle swarm optimization(QPSO)algorithm is a novel algorithm upgrading of PSO algorithm.QPSO algorithm uses the wave function to represent the state of the particle in quantum space.To a certain extent,QPSO algorithm expands the search space of the particle and keeps the diversity of population,when compared with the PSO algorithm.It has better global convergence.The particle still falls into local optimum point due to the decrease of population diversity at the later iteration.In order to solve this problem,this thesis puts forward a series of improved QPSO algorithms respectively for single-objective and multi-objective optimization problems.The new algorithm can enhance population diversity and promote global optimization capability.Then it is applied to the parameter selection of related problems,the multi-objective load forecasting and economic dispatch of power system.The main content of this thesis can be outlined as follows:1.In terms of single-objective optimization,the research has been conducted on search strategy and parallel implementation.Four kinds of improved QPSO algorithms have been proposed in terms of search strategy: improved QPSO algorithm is based on differential evolution,harmony search,black hole and Levy flight mechanism.For improved QPSO algorithm based on differential evolution,we use differential ideas to improve the particle position equation.For improved QPSO algorithm based on harmony search,we use harmony search to eliminate the local optimum point individual continuously.For the two improved QPSO algorithms based on black hole search and Levy flight mechanism,the position equation of the particle are improved.The numerical experiments are used to analyze the strengths and weaknesses of four algorithms.The results show that improved QPSO algorithm based on Levy flight has greater efficiency and better performance than other algorithms.In terms of parallel implementation,a parallel QPSO algorithm with boundary mutation is proposed.The boundary mutation QPSO algorithm simulates particles search process on compute unified device architecture(CUDA)platform with multiple dense computing cores.The results show that this method reduces the optimization time.2.In terms of multi-objective optimization,the parameter selection of multi-objective QPSO algorithm has been analyzed.This thesis proposes a new hybrid multi-objective QPSO algorithm.Firstly,the influence of parameter selection on the performance of multi-objective QPSO algorithm is analyzed by numerical experiments.Secondly,the improved four single-objective strategies are applied to solve unconstrained multi-objective optimization.We verify the strengths and weaknesses of each strategy for unconstrained multi-objective function by numerical experiments.It is concluded that improved QPSO algorithm based on Levy flight mechanism is better than the other three algorithms in terms of the Pareto optimal front distribution.Finally,a new dynamic threshold constraint handling technique is introduced for constrained multi-objective optimization in this thesis.Then adaptive clone rank and dynamic vicinity list multi-objective optimization algorithm combined with QPSO algorithm is proposed.The new algorithm is called hybrid multi-objective optimization algorithm.We also present a new metric called the distribution metric to depict the diversity and distribution of the Pareto optimal front in this thesis.The effectivity of the new algorithm has been proved by simulation experiments.3.The improved single-objective QPSO algorithm is applied to parameter selection of Tikhonov regularization problem and total variation image denoising.The improved QPSO algorithm based on Levy flight is applied to select the Tikhonov regularization parameters.Experimental results show that the proposed algorithm has better performance than other algorithms.For the parameter selection of total variation image denoising,a dynamic parameter selection method is proposed by QPSO algorithm.The experiment results of several standard test images demonstrate that the dynamic parameter selection has better performance compared with the fixed parameters.4.The improved multi-objective QPSO algorithm is applied to solve power load forecasting and economic dispatch problem.Aiming at load forecasting,we first built a multi-objective mathematical model of short-term load forecasting model in this thesis.Then the appropriate data is chosen by discrete particle swarm optimization algorithm.Finally,the improved multi-objective QPSO algorithm is applied to train the diagonal recurrent neural network and multi-objective load forecasting.The numerical experiment shows that the improved algorithm has higher precision of prediction on the load forecasting.Aiming at economic dispatch problem,the mathematical model by consideration of static voltage stability and carbon emissions is given.Then improved multi-objective QPSO algorithm is employed to solve economic dispatch problem.The optimal scheduling scheme is obtained through simulation experiments.