A Study Of Adaptive Memetic Algorithm Based On Particle Swarm Optimization

Memetic algorithm is an effective evolutionary algorithm, which is originally seen as animproved genetic algorithm. With the deepening of the researches, Memetic algorithm hasbeen developed into the framework of the optimization algorithm consisting of global search(Global Search LS) and local search (Local SEARCH LS) strategies. Under this framework,the uses of different search strategies can constitute different Memetic algorithms.Particle Swarm Optimization (PSO) is a random search algorithm which is the analog ofthe biological activities and swarm intelligence in the nature. The idea of Simulated Annealing(SA) derives from the principle of physical annealing. This paper proposes a Memeticalgorithm based on the improved PSO algorithm and SA algorithm (PSO-Memetic PMemetic)by the researches and analysis of Particle Swarm Optimization algorithm and SimulatedAnnealing algorithm for the defects of the existing Memetic algorithm, such as slowconvergence, tending to fall into local extreme values for the comparison with PSO. Theexperimental results show that the PMemetic algorithm improves the global search capability,the convergence speed and the solution accuracy. Experimental results demonstrate that thealgorithm can get the better optimal path. Finally, the algorithm is applied to the intelligent testpaper generation System, and good results are obtained.The main tasks and innovative points are summarized as follows:(1) For the different binding modes of the GS strategy and the LS strategy, twoframework of Memetic algorithm are proposed, which are Sequential Execution PMemetic(S-PMemetic) algorithm and Alternative Execution PMemetic (A-PMemetic) algorithm.S-PMemetic algorithm performs the GS strategy first. If the optimal solution is obtained, theresult is outputted and the algorithm is ended. Or, local searches are performed for theresulting new groups. A-PMemetic algorithm performs the GS and LS strategy alternately,which evolve simultaneously with the implementation of the algorithm. The experimentalresults show that the performance of S-PMemetic algorithm and A-PMemetic algorithm arebetter than PSO. In solving the single and simple functions, efficiency of S-PMemeticalgorithm is higher than that of A-PMemetic algorithm, but the performance of A-PMemetic issuperior to S-PMemetic algorithm in solving multimodal and complexity functions.(2) An improved PSO with the dynamic neighborhood structure is proposed, which isfrom the local decision domain idea of glowworm swarm optimization algorithm. In theglowworm swarm optimization algorithm, the glowworm would be attract by brighterneighbor and move to it. These movements depend only on the local information of glowworm(variable neighborhood range) and interaction between glowworm and neighbors of its choice.Groups are divided into disjoint subgroups, which can make the given multi-peak functionconverges to a multiple optimal solutions. Because each glowworm can be abstracted as a particle, so it is convenient of introducing the idea to PSO. Using the improved PSO as the GSstrategy of S-PMemetic algorithm and A-PMemetic algorithm, can effectively enhance theexecution speed and the precision of solutions.(3) The A-PMemetic algorithm is applied to the intelligent test paper generation problemwith multiple constraints. Teaching management system includes teachers, students andadministrators module. One of the core functions of the system is the interaction betweenteachers and students; another is the intelligent test paper generation function. Intelligent testpaper generation function can provide high quality papers for teachers; enable teachers toonline assignments and online exams. The experimental analysis indicates that the quality oftest paper by the A-PMemetic algorithm is significantly higher than that of PSO; the method iseffective, feasible and practical for test paper generation.