基于PCA-RBF神经网络的浮选过程软测量建模

以浮选过程为研究对象,提出基于主元分析与RBF神经网络相结合的经济技术指标软测量模型,该模型依据工艺机理和经验知识对过程变量进行初选,采用主元分析方法对高维输入向量进行降维化简和辅助变量选择;采用新型混合递推算法对RBF神经网络参数进行优化。该算法包括修正网络中心的自适应聚类的简化型次胜者受罚竞争学习算法和修正网络权值的带遗忘因子的递推最小二乘算法。混合学习算法提高了网络参数辨识的收敛速度。仿真结果表明,软测量模型能很好地实现浮选过程经济技术指标的全局预测。

Soft-Sensor Modeling of Flotation Process Based on PCA-RBFNN

The quality index soft-sensor model of flotation process is proposed based on principal compo- nent analysis(PCA)and radial basis function neural network(RBFNN).Firstly,the process prior knowledge and PCA method are used to deduce the input dimension for RBFNN and predigest model complexity,then a new hybrid recursive algorithm of RBFNN is developed.The algorithm includes sim- plified rival penalized competitive learning method (SRPCL) to make an adaptive clustering of network input pattern and recursive least squares method(LSM)with the forgetting factor,thus updating net- work weights.Simulation results show that the inference estimation model has high predictive accuracy and meets control requirements of the flotation.