基于移动窗口和粒子群寻优的集成偏最小二乘改进算法  

Improved ensemble partial least-squares algorithm based on moving-windows and particle swarm optimization

作　　者：马仕强 任佳[1] 赵梦恩 MA Shiqiang;REN Jia;ZHAO Mengen(Faculty of Mechanical Engineering and Automation,Zhejiang Sci-Tech University,Hangzhou 310018,China)

机构地区：[1]浙江理工大学机械与自动控制学院,杭州310018

出　　处：《浙江理工大学学报（自然科学版）》2018年第4期453-460,共8页Journal of Zhejiang Sci-Tech University(Natural Sciences)

基　　金：国家自然科学基金项目(61203177);浙江省自然科学基金项目(LY17F030024)

摘　　要：为了克服传统偏最小二乘算法对时序非线性数据拟合效果差的问题,提出了结合移动窗口技术和粒子群算法的集成偏最小二乘算法(EMWPLS_PSO)。该算法通过移动窗口判定时序数据状态突变时刻以对原始数据集进行数据划分,添加冗余检查步骤,简化模型结构,同时引入粒子群算法对关键参数寻优,提高了模型性能。采用工业数据集Debutanizer_data验证算法,结果表明:该算法在处理时序、非线性数据时具有更高的精准度和更好的稳定性。这表明基于EMWPLS_PSO的软测量建模算法在工业领域的实用性和可操作性。In order to overcome the poor fitting effect of traditional partial least squares（PLS）algorithm pn nonlinear data,an improved algorithm based on moving window technique and particle swarm optimization（PSO）EMWPLS_PSO was proposed.The moving window was used to determine the mutation time of the time series data so as to divide the original data set.Besides,the redundant checking steps were added to simplify the model structure,and the PSO was introduced to optimize the key parameters and improve the model performance.The algorithm in the paper was proven by testing an industrial data set,Debutanizer_data.The testing result shows that the algorithm is more accurate and stable in processing time series and nonlinear data.It also proves the soft measurement modeling algorithm based on EMWPLS_PSO has practicability and operability in the industry field.

关 键 词：软测量 偏最小二乘 局部加权 移动窗口 粒子群算法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]