基于MFOA优化BP神经网络的磨矿粒度软测量  

Soft Sensing of Grinding Size by Optimized BP Neural Network Based on MFOA

作　　者：杨刚 王建民 

机构地区：[1]华北理工大学电气工程学院,河北唐山市063000

出　　处：《矿业研究与开发》2018年第2期101-105,共5页Mining Research and Development

摘　　要：针对磨矿过程中磨矿粒度难以在线实时检测的问题,提出了一种基于混沌粒子群(CPSO)改进果蝇算法(FOA)优化BP神经网络的方法,建立了磨矿粒度软测量模型。利用混沌搜索的遍历性和对初值的敏感性来提高FOA初始种群的多样性;为了减少适应度函数值更新过程中的盲目搜索,引入了粒子群算法(PSO)。然后利用改进后果蝇优化算法(MFOA)良好的全局寻优能力,自适应地调整BP神经网络的权值和阈值,提高了BP网络的收敛性能和测量精度。选取球磨机给矿量、给水量、磨机电流、分级机溢流浓度和螺旋分级机电流为辅助变量,构建MFOA-BP磨矿粒度软测量模型。研究表明,所构建的MFOA-BP模型鲁棒性强、测量精度较高。Aiming at the problem that it is difficult to achieve reabtime online detection of grinding size, a method that using fruit fly optimization algorithm （FOA） improved by chaotic and particle swarm optimization （CPSO） to optimize BP neural network was proposed. And a soft-sensing model of grinding size was established. The ergodicity of chaotic search and sensitivi- ty to initial values were used to improve the diversity of FOA initial population. In order to reduce the blind search in the process of updating fitness value, the particle swarm optimization （PSO） was introduced. Then, good global optimization abili- ty of MFOA was used to adjust the weight and threshold of BP neural network, which improved the convergence performance and measurement accuracy of BP network. At last, the feeding quantity of ore, feed water, mill current, overflow density and current of spire classifier were selceted as auxiliary variables to establish MFOA-BP soft-sensing model of grinding size. The re- search showed that the constructed MFOA-BP model had strong robustness and high accuracy of measurement.

关 键 词：磨矿粒度 软测量模型 混沌搜索 粒子群算法 果蝇算法 BP神经网络 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TD921.4[自动化与计算机技术—控制科学与工程]