Rough集理论在神经网络材料性能参数实时识别中的应用

Application of the rough sets in identification of material property parameter via neural network

运用神经网络技术实现材料性能参数的实时识别是智能化拉深的重要研究课题。由于训练样本数据的冗余使得BP网络收敛精度差、速度慢,直接影响到网络的识别精度。运用Rough集理论强大的数据分类简约功能,能够去掉多余属性的样本数据,从而优化了神经网络的拓扑结构。经过试验验证优化后的网络不仅收敛速度快、精度得到极大提高,而且网络预测相对误差精度都在6%以下。

The real-time identification via neural network is an important subject in intellectual deep drawing of sheet metal. Because of the redundancy of training data, it makes the convergence of BP neural network slow and imprecise. Using the data reduction and classify function of rough sets, the training data of surplus attrihute can be deleted and the structure of the neural network can be optimized. Experiments prove that the optimized network converges more quickly and accurately. The relative prediction error precisions are all below 6%.