BP神经网络样本数据预处理应用研究

提出一种新的线性预处理输入数据的方法，即通过线性运算将样本数据的各个字段值统一到同一个数量级，然后结合数值归一化的方法将数据运用到神经网络。在基于信令的漫游用户实时信用度测评及欠费风险超前控制系统中，使用统一字段值的数量级的方法预处理样本数据取得了很好的预测效果。由此，在模式识别和预测领域，统一样本数据的各个字段值的数量级后再进行网络训练可以取得更好的训练效果。

Application Research of Preprocess in BP Neural Network Sample Data

A linear preprocess of input data was put forward which changes the eigenvalues magnitude of sample data combining and combines data normalization. The method proposed has a good performance in prediction system of real-time credit evaluation for roaming users based on signaling and control of arrears participated by author. Therefore, In the field of pattern recognition and forecasting, better training results can be gotten after unifying field values of each sample.