The Hierarchical Genetic Algorithm With Improved Penalty Function And Its Application In Structural Optimal Design

For a general structural optimization problem,the design variables are usually discrete,some constraint conditions are not easy to be explicitely formulated,and the objective function is non-derivative.Therefore,traditional methods of structural optimization cannot solve such problems very well.Genetic algorithm(GA)is an intelligent algorithm,which has good adaptability and is not sensitive to the complexity of the problem,and does not need to explicitly formulate the constraints with design variables.Therefore,it can overcome some disadvantages of traditional optimization algorithm.In this thesis,the basic principle of GA was introduced.Then,it was improved according to the characteristics of the structure optimization.Compared to the basic GA,the improved algorithm can avoid to be trapped by local minimum and can convergence more stable.In addition the improved GA makes more suitable punishment according to the degree of individual deviation.In this way,the non-feasible solution space can be treated better,and the potential non-feasible solution individuals remain reproductive opportunity.The improved GA maintains the diversity of individual population and increases the efficiency of searching.By classifying and sequencing the current population,excellent individuals are given the chance to reproduce and avoid falling into local optimal solutions.In this paper,we test the algorithm through three classic examples of De Jong,Shubert and Griewank,the results show that the proposed method has obvious advantages.The proposed method is used to optimize two classical structural models: a 17-bar plannar truss structure and a 72-bar spacial truss structure.The results show that the improved GA has better optimization performance,and it can be extended to the optimization design of other structural types.The genetic algorithm was mainly used to optimize simple structure in the past due to the massive computation amounts from reanalyzing the finite element model for each individual in each iteration cycle.It is required to enhance the computational efficiency in both sides of hardware and software.In this thesis,the parallel computing platform is establised and the cycle mechanism of genetic algorithm is optimized to accelerate the computation and avoild program collapse.The results of the optimization of the large cable mast(1,164 bars)and the super-tall frame structure(5,980 bars)show that the improved GA can be applied in the optimal design of complex structures.