摘要
针对最小二乘支持向量机(LS-SVM)缺少支持向量所具有的稀疏性和模型参数值难以选择的问题,提出利用马氏距离进行样本相似程度分析,去除集中部分样本,以恢复最小二乘支持向量机的稀疏性的方法。同时,采用k-折交叉验证误差作为学习目标的粒子群优化算法来选取模型参数,并利用改进算法建立了精馏产品浓度的软测量模型。通过仿真验证了改进算法的有效性。结果表明模型精度较高,泛化能力强,满足工业测量要求。
Aiming at the demerits of the least square support vector machine,such as losing the sparseness of support vector,and difficulty in selecting parameters of model,it is proposed that by using Mahalanobis distance to analyze the similarities among samples,for recover the sparseness of the least square vector machine.In addition,by adopting particle swarm optimization algorithm that with k-fold cross-validation error as learning object to select parameters of the model,and the improved algorithm is used to establish the soft sensing model for concentration of the product of distillation.The simulation verifies the effectiveness of the improved algorithm,and the result of research shows the model is accurate,and good for generalization to meet industrial measuring requirements.
出处
《自动化仪表》
CAS
北大核心
2011年第5期39-41,45,共4页
Process Automation Instrumentation
关键词
最小二乘支持向量机
粒子群优化算法
软测量
建模
Least square support vector machine（LS-SVM） Particle swarm optimization algorithm Soft sensing Modeling
