Research And Improvement Of Particle Swarm Optimization

Optimization problems are not only common in fields such as technological development and engineering applications,but their complexity is also increasing with various needs.Particle swarm optimization algorithm has received widespread attention due to its simple and efficient advantages and has been successfully applied to solve practical optimization problems.However,the capability of particle swarm optimization algorithm to resist premature traps is still insufficient.There are many problems such as the processing method and response strategy are very single,and the relationship between the particle's own state and the update strategy is ignored.In view of the above problems,this thesis proposes the adaptive hierarchical update particle swarm optimization algorithm with a multi-choice comprehensive learning strategy(AHPSO)to enhance the resistance of particles trapped into local optima.According to the tournament mechanism,the algorithm divides the population into two subpopulations that undertake global exploration and local exploitation task,respectively.Under the overall planning and coordination of “Local Optimum Early Warning”,particles are updated according to the best strategy.The main work and innovation involved in this thesis are shown as following:(1)Aiming at the problem of local optimal lacking appropriate monitoring methods and evaluation methods,a new concept is introduced as “Local Optimum Early Warning”,which can be employed to reflect the current degree of risk and urgency of particles being trapped in local optima.Based on the calculated value of the “Local Optimum Early Warning”,the most suitable update formulas and learning paradigms are matched for the particle,and the dynamic balance between a particle's own state and the update strategy is realized.(2)Aiming at the problem of algorithm lacking a local optimal-oriented particle update formula,the proposed adaptive hierarchical update method develops two-layer and three-layer update formulas for the global exploration subpopulation and the local exploitation subpopulation,respectively.It can help particles to optimize with a greater convergence rate and maintain a higher level of population diversity.(3)Aiming at the phenomenon that the effect of learning paradigm guiding the particles to get rid of the local optimum is poor,a multi-choice comprehensive learning strategy is also developed to select the most suitable construction rule from the weighted synthesis sub-strategy and the mean evolution sub-strategy for learning based on the current degree.The strategy then efficiently utilizes the selected superior information providers to generate the most suitable paradigm that guides the trajectory of particles,according to a reasonable dimensional segmentation method.Eighteen benchmark functions and one real-world optimization problem are employed to evaluate the AHPSO against eight typical PSO variants.According to the experimental results,the AHPSO outperformed other methods in solving different types of functions by yielding high solution accuracy and high convergence speed.