基于RBFNN模型明胶浓度在线测量

摘要

For improving the measurement efficiency of the gelatin concentration, the paper presents a soft sensor multi-model based on radialrnbasis function neural network (RBFNN). It chose time temperature and density as instrumental variables. The obtained datas were clustered by GK clustering algorithm. The RBFNN has been constructed by the RBF network module of the NeuroSolution software. Each subclass gathered was established by RBFNN model. The outputs corresponding input variables of the subclasses were as the the system model final one. The simulation result shows MSE is 0. 000 824 and the accuracy of the multi-model RBFNN is higher than the single one.%为提高国内明胶企业检测明胶浓度效率,提出一种基于径向基函数人工神经网络(RBFNN)软测量多模型.选定时间、温度和比重作为辅助变量,用GK聚类算法对所采集的数据进行聚类,使用NeuroSolution软件中的RBF模块组成RBFNN,将所聚子类数据输入该模型进行训练,用与输入变量对应的子模型的输出作为系统最终输出.仿真结果表明该建模方法均方根误差为0.000 824,与相同辅助变量单RBFNN模型相比精度有了很大提高.