对比了三种不同神经网络模型的生成方式:传统神经网络生成模型,遗传算法训练神经网络模型,以及在第二种方式训练参数的基础上,再使用传统神经网络优化生成模型.论文使用上述三种方法对代表性股票和商品价格进行拟合并预测,通过预测结果准确性和稳定性的比较发现:引入遗传算法后的神经网络在样本内的拟合误差有所降低,而第三种方法在样本外有最低的预测误差和最优稳定性. Three different neural network models were compared: the traditional neural network generation model, the genetic algorithm training neural network model, and the second training parameter, and then the traditional neural network was used to optimize the generation model.This paper uses the above three According to the comparison between the accuracy and the stability of the prediction results, the fitting error of the neural network after introducing the genetic algorithm in the sample is reduced, while the third method The lowest prediction error and best stability outside the sample.