Study On The Improvement Of Genetic Algorithms And Their Application In The Engineering Optimization

Most of the present single objective genetic algorithms are with problems of prematureconvergence, genetic shift and the diversity is not well maintained. As for themulti-objective genetic algorithms, the distribution and convergence of the final solutionsis always not satisfying. In consideration of this, four improved genetic algorithms areproposed so as to enhance the property of the algorithm.The first proposed algorithm is called the new crowding self-adaptive genetic algorithm(CSAGA). This algorithm introduces a crossover and mutation self-adaptive adjustmentmethod based on Sigmoid function to improve the inferior individuals and avoidevolutionary stagnation and local convergence; a new crowding mechanism based on themost similar individuals is adopted so as to maintain the diversity of the population andensure the stability of the algorithm. The tests on the typical test functions show that theCSAGA is of excellent convergence speed, stable convergence success rate, higherconvergence accuracy and outstanding capacity for maintaining diversity.The second improved algorithm is a new crowding niches genetic algorithm based onthe most similar individuals (MSICNGA) for solving the multimodal functions. It includescrowding system based on the most similar individuals and crowding error repairingsystem. The former system maintains the population diversity by recurrent squeezing outthe inferior one of the two most similar individuals in the population, while the lattersystem protect the optimal solution from being deleted by saving the superior individualsof the removed ones during the crowding process. The multimodal function testsdemonstrate that MSICNGA performs well in maintaining population diversity. It caneffortlessly search many optimal solutions and can avoid genetic shift effectively.The third one is a multi-objective genetic algorithm based on cloning mechanism(NMGA). For one thing, the superior individuals are picked out by double binarytournament selection, and then these individuals are taken as a whole to be cloned. Itgreatly enhances the information transmission between the parents and the offsprings, andimproves population convergence and diversity effectively. For another, a compoundcrossover operator is designed, which makes the crossover coefficient of the individualshave a dynamic connection with their ranks. The individuals with lower ranks are reserved, while those with higher ranks are crossed. The simultaneous calculation on NSGAII andNCMA shows that NMGA clearly outperforms NSGAII in terms of convergence anddiversity.The fourth algorithm is a differential cellular multi-objective genetic algorithm (DECell)which introduces differential evolution strategy into canonical cellular multi-objectivegenetic algorithm and adopts commutant mutation operator. It ensures the uniformity andwide coverage area of the gained Pareto front, meanwhile, the Pareto front graduallyapproximates to the optimal front. Test on benchmarks indicates the algorithm’s betterconvergence, diversity and expansibility in solving the multi-restrains, multivariable andnon-linear models.In order to evaluate the efficiency of the algorithms in solving practical engineeringproblems, these proposed algorithms have been applied to engineering practices. CSAGAis used to compute the single-objective optimization model of the workshop equipmentlayout. The result shows that CSAGA performs effectively in computing this model andthe solutions gained are of high quality; as for DECell, when computing themulti-objective workshop layout model, it can obtain more better Pareto fronts; whileoptimizing the ultra-precise vibration isolation platform integrated optimized design model,the Pareto fronts gained by NCMA distribute more uniform and it works effectively inoptimizing the acceleration, dynamic displacement and dynamic deflection; DECell isapplied to brake multi-objective optimization design, the result indicates that DECell canget Pareto fronts that distribute more uniform in solution space, and DECell can find outbetter limit point for each objective.