基于SOM和RBF网络的软测量方法研究

Research and Application of Soft Measurement Method Based on SOM and RBF Network

为了解决工业生产过程中许多重要的参数无法精确测量或者实时测量的问题,提出一种基于自组织特征映射(SOM)神经网络和径向基函数(RBF)神经网络结合构建网络模型的预测方法;其中,RBF神经网络作为基础网络实现从输入层到输出层的线性映射,得出预测输出;SOM神经网络作为聚类网络对输入样本进行自组织分类,将分类中心及其对应的权值向量作为RBF神经网络径向基函数的中心;以钕铁硼氢粉碎过程优化控制为例,建立了合金氢含量的检测模型,并与RBF神经网络检测模型进行了对比;仿真结果表明该混合网络检测模型检测精度高,泛化能力强,证实了该方法的有效性。

In order to solve the problem that many important process parameters can not be accurately measured or real time measure ment in industrial, proposed a forecasting method of network model based on Self--Organizing Feature Maps （SOM） neural network and Ra dial Basis Function （RBF） neural network, for this model, RBF neural network as the basis network to achieve linear mapping from the input layer to the output layer for obtaining predictive output; SOM neural network as clustering network to classify the input samples for self-organization, and its classification center and corresponding weight vector are the center of the radial basis function for RBF neural network. The process optimization control of NdFeB hydrogen pulverization is taken as an example to build the detection model of hydrogen content of alloy, and we compare this with the detection model by using the RBF neural network. The results of simulation indicate that the hybrid net work detection model has high precision and generalization ability, and confirming the effectiveness of the method.