基于偏最小二乘及最小二乘支持向量机的人工加糙渠道糙率预测模型

影响渠道糙率的因素相当复杂,且因素间又存在一定的相关关系。为取得更为精确的糙率预测效果,采用偏最小二乘(PLS)法对影响人工加糙渠道糙率的因素进行分析,提取影响自变量的重要成分,结合最小二乘支持向量机(LSSVM)建立了人工加糙渠道糙率预测模型。结合实例,通过对某人工加糙渠道相关试验数据进行PLS-LSSVM模型的训练及预测,并将预测结果与单独使用PLS、LSSVM及公式法的预测结果进行对比,其结果显示:基于PLSLSSVM模型的预测平均绝对百分比误差MAPE为1.38%,均方根误差RMSE为2.24×10~(-4),预测精度均优于PLS、LSSVM及公式法的预测结果。结果表明,将PLS与LSSVM相结合的PLS-LSSVM模型,综合了PLS与LSSVM各自的优势,应用PLS-LSSVM模型可有效进行人工加糙渠道糙率的预测。

Roughness prediction model for artificially roughened channel based on partial least squareand least square support vector machine

The fact ors that affect the roug hness o f a channel are quite complex , and there is a cer tain co rr elatio n between the fac2to rs. In o rder to o bt ain a more accurate prediction of t he r oughness, we used the pa rtial least squares ( PLS) method to analyzethe facto rs that affect the r oughness o f ar tificially r oughened channels, and we extr acted t he import ant compo nents that affectthe independent v ariables. Then w e established the ro ug hness predictio n model for artif icially ro ug hened channels based on leastsquar e suppo rt v ect or machine ( LSSVM) . We used the ex per imental data of an ar tificially roug hened channel fo r training andpr ediction of the PLS2LSSVM mo del, and compar ed the pr edict ion results w ith the pr edict ion results of PLS, LSSVM, and for2mula metho ds. T he results showed that the mean absolute percentag e er ro r (MA PE) o f predictio n based o n PLS2LSSVM modelwas 11 38%, and t he r oot mean square er ro r ( RMSE) was 21 24 @ 1024 . Its predictio n accuracy w as better than t hat o f t he PLS,LSSVM, and fo rmula methods. The results show ed that the PLS2LSSVM mo del w hich combines PLS and LSSVM can int eg rat ethe advantages of PLS and LSSVM. PLS2LSSVM model can effect ively pr edict the roug hness of ar tificially roug hened channels.