本文提出一种新颖的基于实数编码的改进遗传算法用于神经网络的训练 ,该算法采用基于代沟最小的代选择模型 ,选用 BLX-α混合交叉算子 ,算法简单、鲁棒性强 ,优化兆效率高 .并将该神经网络应用于非线性系统预测 ,仿真结果证实该方法是快速有效的 In this paper, a novel genetic algorithm based on real number coding is proposed for neural network training. This algorithm uses the generation selection model based on the smallest generation gap and selects BLX-α hybrid crossover operator. The algorithm is simple, robust and optimized High efficiency and the neural network is applied to the nonlinear system prediction, the simulation results confirm that the method is fast and effective