EMD-SVM非线性组合模型对高炉铁水含硅量的预测

Prediction of silicon content in hot metal based on EMD-SVM nonlinear combined model

提出了一种基于经验模式分解(EMD)和支持向量机(SVM)的非线性组合模型的预测方法.该方法运用EMD将原始铁水含硅量的时间序列分解成若干个频率不同的平稳分量,分解后的分量突出了原序列的局部特征.通过Lempel-Ziv复杂度分析选用不同的核函数,并利用10-fold交叉检验方法取定相应的参数,从而对各个分量构建不同的支持向量机模型,并对各分量进行预测.仿真结果表明,EMD-SVM非线性组合模型预测命中率达到90%.

In the blast furnace （BF） ironmaking process, the silicon content in hot metal which reflects the thermal state of BF is an important index. In order to predict the silicon content in hot metal effectively and level up the forecasting accuracy, a combined model based on Empirical Mode Decomposition （EMD） and Support Vector Machine （SVM） is proposed. Firstly, the time series data of silicon content in hot metal are decomposed into a series of stationary intrinsic mode functions （IMFs） in different scale space via the EMD sifting procedure. The local features of original time series data are prominent in the IMFs. Secondly, based on the analysis of Lemple-Ziv complexity and 10-fold cross validation, the right kernel functions and their parameters are chosen to build different SVMs respectively to predict each IMF. Finally, the predicted results of all IMFs are reconstructed to obtain the final predicted result. The result shows that the prediction is successful and the hit rate increased to 90 %.