摘要
遗传算法具有搜索全局最优解的能力,并且有很强的鲁棒性,而BP算法具有很好的泛化能力和非线性映射能力,基于两种算法的特点,设计了一种GA-BP算法,该算法将遗传算法应用到神经网络中权值和阈值的优化中,将最优解的分布范围缩小,然后通过BP算法进行再次优化和精确求解,以防止神经网络陷入局部极小点,从而达到加速收敛、减少训练次数的目的;并且通过对比实验给出该算法的可行性和有效性分析,进一步验证了该算法在收敛速度和误差精度上的优越性。
Genetic algorithm has ability of searching the global optimal solution and strong robustness, and BP algorithm has better ability of generalization and nonlinear mapping, according to the characteristics of two algorithms, a GA-BP algorithm is designed. In this algorithm, genetic algorithm is applied to the optimization of Neural Network weights and thresholds to reduce distribution range of optimal solution,then using BP algorithm for re-optimization and exact solving in order to prevent the Neural Network into a local minima, thus achieving the purpose of accelerating convergence and reducing training times;the analysis on feasibility and effectiveness of this algorithm is given by comparative experiments, it further verifies the superiority of this algorithm in convergence rate and error precision.
出处
《大众科技》
2014年第4期24-26,37,共4页
Popular Science & Technology
关键词
遗传算法
神经网络
种群
变异
适应度
Genetic algorithms
neural network
population
mutation
fitness