摘要
介绍了基于BP(Back Propagation)神经网络建立市场销售预测模型的设计方法。针对BP算法的不足之处提出了改进措施:给网络参数赋初值;对原输入信息表达空间进行归一化;修正权重系数;将所有数据作为样本输入网络进行训练等。实验预测结果表明:BP算法适用于规则不可知的预测内容,它比常规方法具有更高的逼近精度和更好的预测能力。
A design way of setting up market forecasting model is introduced basing on Back-Propagation neural network. The improving method is put forward aiming at defect of BP algorithm: assigning initial value to network parameter; normalizing the expression space of original input message; revising the weighing coefficient; inputting all the datas to network for training as sample. The test forecasting result shows: BP algorithm is suitable for forecasting on some irregular questions. It has higher approximation precision and better forecasting ability than regular way.
关键词
BP神经网络
销售
预测
BP neural network
market
forecasting