Research On Global Annealing Evolutionary Model And Distributed Computing Method

Evolutionary computing is an intelligent search technology with strong applicability,which has been successfully applied to the optimization of a large number of complex problems.Analog circuit evolutionary design is one of the most important applications of evolutionary computing.It is characterized by bottom-up circuit design.For top-down conventional design based on empirical knowledge and design rules,evolutionary circuit design provides a novel approach to automated design.Analog circuit evolutionary design can use evolutionary algorithms to automatically search and optimize the structure and parameters of circuits,and can also be used as a supplementary method for parameter optimization in conventional circuit design.The research on analog circuit evolutionary design has important academic significance and application exploration value.At present,the search performance and computational time consumption of evolutionary algorithms in the affordable time range are the core issues in the evolutionary circuit design,which directly affects the practical application of evolutionary circuit design.In this thesis,the basic methods,key technologies and research status of analog circuit evolutionary design are analyzed and summarized.The selection pressure of genetic algorithm and the influence of mobility parameters of distributed genetic algorithm on the convergence performance of evolutionary model are emphatically studied.The role of task allocation algorithm in distributed computing in reducing system time-consuming is also discussed.The main research work is as follows:Based on the selection pressure of genetic algorithm,considering the influence of selection pressure on population diversity and convergence performance,a new global simulated annealing evolutionary model is proposed by comparing and analyzing the effects of simulated annealing algorithm(SA)and genetic algorithm(GA).In this model,the temperature parameter T in SA is introduced into the basic framework of GA,and the selection pressure of the algorithm is dynamically adjusted by the decreasing trend of temperature and the degree of population evolution,and the performance of the evolutionary design model is improved by controlling the selection pressure.The experimental results show that the model can achieve better design efficiency when dealing with circuit synthesis problems.The mobility of island model in distributed genetic algorithm is an important parameter affecting the performance of island model.This thesis analyses the effect of mobility settings on the performance of the algorithm,and proposes a stage dynamic mobility strategy.This strategy makes the mobility rate increase gradually with the deepening of the evolution process in a follow-up manner,which can better deal with the relationship between population diversity and convergence in the later stage of evolution.Compared with the island model algorithm with fixed mobility rate,this strategy can effectively improve the algorithmic convergence performance.A dynamic task allocation strategy for circuit evolutionary design is designed and implemented.Aiming at the Master/S lave architecture in distributed evolutionary computing,the strategy takes into account the heterogeneity of actual computing devices and the uncertainty of circuit simulation time consumption.The dynamic task allocation strategy is characterized by allowing Slave to monitor its idle time and actively request sub-tasks from Master.The "sub-task combination" and "sub-task retransmit" mechanisms are added to the strategy,which reduces the communication time between Master and Slave and ensures the fault tolerance performance of the system.The experimental results show that the strategy can reasonably and efficiently guarantee the load balancing of computing devices,reduce the evolutionary time,and make the distributed evolutionary system have good fault tolerance and scalability.In a word,this thesis studies the selection pressure of genetic algorithm,mobility rate of island model and load balancing of distributed evolution,which improves the convergence performance of evolutionary design model and reduces the time consumption of evolutionary design.