基于最优变异因子的遗传算法在ANN训练中的应用

Best mutation coefficient genetic algorithm for ANN training

为解决人工神经网络训练中陷入局部极小值问题 ,对遗传算法中的变异模型作了分析和改进 ,采用了一维搜索方法以确定最优变异因子 ,首先由进退法确定最优变异因子存在的区间 ,然后运用黄金分割法以确定最优变异因子。结合一个实际算例 ,对最优变异因子和固定变异因子的应用效果进行了比较 ,结果表明基于最优变异因子的遗传算法能够更有效地克服局部极小点 。

Local minima encountered in the training of artificial neural network (ANN) can be overcome using a mutation model of the genetic algorithm (GA) with the best mutation coefficient. A one-dimensional search approach is utilized to achieved the best mutation coefficient. The limits of the best mutation coefficient are determined using a scanning process and then the golden section method is used to search for the best mutation coefficient. A case study of a real petrochemical process is used to compare the results using the best and the constant mutation coefficients. The results show that the improved GA more effectively overcomes local minima, and therefore, accelerates the ANN training.