摘要
针对现有重介质悬浮液密度测量方法易受异常值影响等问题,提出了一种基于随机配置网络(SCN)的重介质悬浮液密度鲁棒软测量方法。基于SCN模型的近似误差来增量配置具有约束随机参数的隐藏节点,确认输入权重及输入偏置;采用具有厚尾特性的Laplace分布建模输出测量噪声,结合期望最大化(EM)算法估计输出权重,提升模型在异常值干扰下的鲁棒性。针对传统SCN监督机制中残差变量易受到负面影响,导致监督机制约束能力减弱的问题,提出了一种基于Laplace分布的增量式鲁棒随机配置网络(IRSCN),通过惩罚权重对SCN的监督机制进行加权,进一步抑制数据异常值对隐藏节点随机参数选择的干扰。实验结果表明:对比同类型鲁棒模型RANSAC−SCN,IRSCN在最大延迟响应、平均延迟响应与训练时间上均有优势,表明IRSCN的实时性更优,计算效率更高;在数据中存在异常值污染的情况下,IRSCN的RMSE平均值显著低于RANSAC−SCN与SCN,其整体误差水平最低;IRSCN在真实数据集上的误差明显小于其余模型,且能够更准确稳定地捕捉到重介质悬浮液密度的真实变化趋势。
To address the problem that existing dense medium suspension density measurement methods are easily affected by outliers,a robust soft measurement method for dense medium suspension density based on Stochastic Configuration Network(SCN)was proposed.Hidden nodes with constrained random parameters were incrementally configured based on the approximation error of the SCN model to determine input weights and biases.The output measurement noise was modeled using a Laplace distribution with heavy-tailed characteristics,and the Expectation-Maximization(EM)algorithm was applied to estimate output weights,thereby enhancing the robustness of the model under outlier interference.To solve the problem that the residual variables in the traditional SCN supervisory mechanism were susceptible to negative effects,which weakened the constraint ability of the supervisory mechanism,an Incremental Robust Stochastic Configuration Network(IRSCN)based on the Laplace distribution was proposed.By applying penalty weights to the supervisory mechanism of SCN,the interference of data outliers with the selection of hidden node random parameters was further suppressed.Experimental results showed that,compared with the robust model RANSAC-SCN,IRSCN had advantages in maximum delay response,average delay response,and training time,indicating better real-time performance and higher computational efficiency.Under the condition of data contamination with outliers,the mean RMSE of IRSCN was significantly lower than that of RANSAC-SCN and SCN,with the lowest overall error level.On real datasets,the error of IRSCN was markedly smaller than that of other models,and it captured the actual variation trend of dense medium suspension density more accurately and stably.
作者
彭启月
楚宁馨
刘小庆
周佳奕
蔺楚仪
PENG Qiyue;CHU Ningxin;LIU Xiaoqing;ZHOU Jiayi;LIN Chuyi(School of Mechanical and Electrical Engineering,China University of Mining and Technology,Xuzhou 221116,China;School of Information and Control Engineering,China University of Mining and Technology,Xuzhou 221116,China)
出处
《工矿自动化》
北大核心
2025年第8期95-101,116,共8页
Journal of Mine Automation
基金
中国博士后面上项目(2023M743776)。
关键词
重介质选煤
悬浮液密度测量
随机配置网络
期望最大化算法
LAPLACE分布
监督机制
dense medium coal preparation
suspension density measurement
Stochastic Configuration Network
Expectation-Maximization algorithm
Laplace distribution
supervisory mechanism
