基于最小二乘支持向量机的氧化铝浓度预测  

Prediction of Alumina Density Based on LSSVM

作　　者：崔桂梅[1] 杨海靳 刘丕亮 于凯[2] Cui Guimei;Yang Haijin;Liu Piliang;Yu Kai(School Of Information Engineering, Inner Mongolia University Of Science ＆ Technology, Baotou 014010, China;School Of Mathematics Physics, Inner Mongolia University Of Science ＆ Technology, Baotou 014010, China)

机构地区：[1]内蒙古科技大学信息工程学院,内蒙古包头014010 [2]内蒙古科技大学数理学院,内蒙古包头014010

出　　处：《系统仿真学报》2018年第5期1844-1849,共6页Journal of System Simulation

基　　金：国家自然科学基金(61164018);内蒙古自治区自然科学基金(2014MS0612)

摘　　要：建立基于种群活性粒子群算法优化最小二乘支持向量机参数的铝电解氧化铝浓度预测模型,根据铝电解生产工艺及历史数据的特点,确定模型的基本输入变量,解决了粒子群优化算法早熟收敛及最小二乘支持向量机参数确定周期长的问题。利用群活性加速度作为多样性测度,当群活性的加速下降时,对粒子的位置和速度分别执行进化操作用以改进标准PSO(Particle Swarm Optimization)算法;与标准的LSSVM(Least Squares Support Vector Machine)方法相比,提出的改进PSO-LSSVM的预测模型有效地提高预测精度且计算速度更快。The prediction model of alumina density based on the PSO algorithm with swarm activity to optimize LSSVM method is built. According to the production process characteristics of aluminum electrolysis and historical data, the input variables of the model is determined. It can solve these problems that Particle Swarm Optimization （PSO） algorithm is with the risk of premature convergence and least square support vector machine is time consuming with parameter selection. The method uses swarm activity as diversity index. When swarm activity is quickened to descend, evolution operation is added to modify the positions or velocities of particles to improve standard PSO algorithm. Study shows that, the improved PSO-LSSVM prediction method has better estimating performance and less computational time than the traditional LSSVM method.

关 键 词：铝电解 氧化铝浓度 最小二乘支持向量机 粒子群算法 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]