Abstract: The flatness defect pattern recognition based on data mining technology was proposed. In order to solve low accuracy of normal BP (Back Propagation) network, a novel data mining algorithm based on hierarchical BP model was presented. The new model with binary tree structure reduced prediction range of each network and adopted several networks for degree elevation. Compared with the normal BP model, the new system precision was improved remarkably. The experimental results show this method can meet the requirements of the producing process.

Key words: data mining, artificial neural network, flatness defect, pattern recognition, hierarchy

摘要： 介绍了建立带钢板形缺陷模式识别的数据挖掘过程。针对普通神经网络识别精度较低的缺陷，提出一种基于分层神经网络进行数据挖掘的新方法。该方法采用二叉树型结构，通过分层来细化预测范围并选用多个神经网络进行递推。实验结果证明了分层神经网络模型比普通神经网络模型的预测精度有较大提高，完全可以满足实际生产需要。

关键词: 数据挖掘, 人工神经网络, 板形缺陷, 模式识别, 分层