An Improved Real-Coded Genetic Algorithm Using the Heuristical Normal Distribution and Direction-Based Crossover

Abstract

A multi-offspring improved real-coded genetic algorithm (MOIRCGA) using the heuristical normal distribution and direction-based crossover (HNDDBX) is proposed to solve constrained optimization problems. Firstly, a HNDDBX operator is proposed. It guarantees the cross-generated offsprings are located near the better individuals in the population. In this way, the HNDDBX operator ensures that there is a great chance of generating better offsprings. Secondly, as iterations increase, the same individuals are likely to appear in the population. Therefore, it is possible that the two parents of participation crossover are the same. Under these circumstances, the crossover operation does not generate new individuals, and therefore does not work. To avoid this problem, the substitution operation is added after the crossover so that there is no duplication of the same individuals in the population. This improves the computational efficiency of MOIRCGA by leading it to quickly converge to the global optimal solution. Finally, aiming at the shortcoming of a single mutation operator which cannot simultaneously take into account local search and global search, a Combinational Mutation method is proposed with both local search and global search. The experimental results with sixteen examples show that the multi-offspring improved real-coded genetic algorithm (MOIRCGA) has fast convergence speed. As an example, the optimization model of the cantilevered beam structure is formulated, and the proposed MOIRCGA is compared to the RCGA in optimizing the parameters of the cantilevered beam structure. The optimization results show that the function value obtained with the proposed MOIRCGA is superior to that of RCGA.