正则参数求解的微分进化算法

研究了正则化方法中正则参数的求解问题,提出了利用微分进化算法获取正则参数.微分进化算法属于全局最优化算法,具有鲁棒性强、收敛速度快、计算精度高的优点.把正则参数的求解问题转化为非线性优化问题,通过保持在解空间不同区域中各个点的搜索,以最大的概率找到问题的全局最优解,同时还利用数值模拟将此方法与广义交叉原理、L-曲线准则、逆最优准则等进行了对比,数值模拟结果表明该方法具有一定的可行性和有效性.

Solving Regularization Parameters Based on Differential Evolution Algorithm

In this paper,we research the problem of solving regularization parameter in the regularization method and put forward a method for solving regularization parameters which is using differential evolution algorithm.The differential evolution algorithm is one kind of global optimization method with strong robustness,rapid convergence and higher accuracy.The problem of solving regularization parameter is translated into nonlinear optimal problem.Each point in different region of solution space is researched. The global optimal solution is found with the greatest probability.Then we compare the method with generalized cross-validation criterion,L-curve method and inverse optimality method through numerical simulation.The results of numerical simulation indicate that this method is feasible and effective.