基于交叉算子的反向梯度优化算法

A Reverse Gradient Optimization Algorithm Based on Crossover Operator

梯度优化算法(Gradient-based optimizer,GBO)是从牛顿法中得到启发,采用梯度搜索规则(GSR)和局部逃逸算子(LEO)进行搜索的一种新算法.针对算法求解精度低、收敛速度慢及容易陷入局部最优的不足,提出了一种基于交叉算子的反向梯度优化算法(MGBO).该算法将每次迭代所得种群中较差的一半个体与当前最优个体进行两两交叉以提高解的精度并增加种群多样性;然后对搜索后所得种群中所有个体进行反向学习操作,使得算法有能力逃离局部最优.在9个基准函数上的实验结果表明,改进算法的收敛速度和求解精度得到较大的提升.

Gradient-based optimizer(GBO)is a new algorithm inspired from Newton's method,which searches for optimal solutions by using gradient search rule(GSR)and local escape operator(LEO).Aiming at the shortcomings of low solution accuracy,slow convergence and easy to fall into local optimum,a reverse gradient optimization algorithm based on crossover operator(Modified Gradient-Based Optimizer,MGBO)is proposed.The proposed algorithm crosses the worse half of the population in each iteration with the current best individual to improve the solution accuracy and increase the population diversity.Then,it performs the opposition-based learning operation for all individuals in the population after the search,and the algorithm has the ability to escape from the local optimum.The experimental results on nine benchmark functions show that the convergence speed and solution accuracy of the improved algorithm are greatly improved.