摘要
根据带钢张力时间序列非平稳、非线性特征,提出一种基于经验模态分解(EMD)和支持向量机(SVM)组合模型的预测方法;首先,应用EMD方法将原始张力序列分解成若干不同频率的平稳分量;其次,根据各分量特征,选用合适的核函数和最佳参数建立不同的SVM回归分析模型,对各分量测试集进行SVM预测;最后将各分量预测序列组合成原始序列的预测值;将EMD-SVM模型用于带钢张力预测,并与ARMA和SVM模型预测结果比较;EMD-SVM模型预测相关度可高达99.93%,而ARMA和SVM模型预测的相关度分别只有88.82%和79.31%,仿真结果表明EMD-SVM模型有较高的预测精度。
According to nonlinear and nonstationary characteristics of time series of strip tension, a prediction method based on Empirical Model Decomposition （EMD） and Support Vector Machine （SVM） combined model is proposed. Firstly, the original tension series are de- composed into a series of stationary intrinsic mode functions in different scale space via EMD sifting procedure. Secondly, according to the change regulation of each of all resulted intrinsic mode function, the kernel functions and right parameter are chosen to build different SVM regression analysis models are built to predict each test IMF. Finally, the predicted results of all IMFs are combined to obtain the predictive value of the original series. This EMD--SVM prediction model is applied to a strip tension series example, and then compared with the AR- MA and SVM prediction results. The correlation of EMD--SVM prediction model is up to 99.93 %, but the correlation of ARMA and SVM prediction are only 88.82% and 79.31%, so the simulation results shows that the EMD--SVM regressio nanalysis model has high prediction accuracy.
出处
《计算机测量与控制》
北大核心
2014年第4期1279-1281,1284,共4页
Computer Measurement &Control
基金
国家自然科学基金项目(61174106)
关键词
带钢张力
经验模态分解
支持向量机
回归分析
预测
strip tension
empirical model decomposition
support vector machine
regression analysis
prediction
