EMD分解和SVM模型在时间序列荷载预测中的应用
发布时间：2018-09-09 18:18
【摘要】：在水利工程中有很多高耸或长大结构,如高坝及其坝顶建筑、工作桥及大跨渡槽等,而风和地震等时间序列荷载对这些结构影响的很大,一些工程因之而受到影响而出现事故,甚至造成灾难。深入研究风与地震等时间序列荷载的特性和规律,并对其进行预测,将有助于结构的振动控制,避免或减轻工程灾害。本文应用经验模态分解及支持向量机模型,研究风速及地震加速度的预测问题,主要内容如下:1)分析了风、地震所具有的不稳定的特点以及应用于预测非平稳数据序列的方法,对传统预测方法的特点和目前应用较多的预测方法进行了比较。2)介绍了基于统计学理论的支持向量机和基于结构风险最小化的最小二乘支持向量机学习方法的原理;经验模态分解方法的基本原理、详细分解步骤及一些问题的解决方法;为了提高模型预测准确性,引入粒子群优化算法,建立预测模型时模型核函数和参数选择问题、具体的模型建立步骤。最后以风速预测和地震加速度预测为例验证这种预测方法的有效性。3)提出了基于经验模态的分解与多步预测的最小二乘支持向量机相结合的方法,对风速的非线性时间序列分析进行了建模预测。首先对风速动态信号加以经验模式的分解,将原信号分解为若干个不同特征尺度(频率)的本征模态函数。再建立多步预测为基础的最小二乘支持向量机预测模型,对不同频带的平稳IMF分量进行预测,将各分量的预测值等权求和得到最终预测值。发现EMD与多步预测的LS-SVM相结合的风速预测方法比单一的LS-SVM预测方法的预测精度更高。在此基础上应用PSO优化算法对模型参数的选择进行优化,得到优化后的预测结果。4)因为地震加速度同样具有非平稳的时间序列的特点,所以,同样地应用基于PSO优化的EMD-LS-SVM预测方法对加速度序列进行建模预测。计算结果验证了这种组合方法在非平稳时间序列预测上的适用性。最后对本论文中的成果进行了分析总结,对今后的研究方向进行了展望。
[Abstract]:In water conservancy projects, there are many tall or long structures, such as high dams and their top buildings, working bridges and long span aqueducts. However, time series loads such as wind and earthquake have a great impact on these structures, and some projects are affected by accidents as a result. Or even disaster. Further study on the characteristics and rules of time series loads such as wind and earthquake, and prediction of them will be helpful to the vibration control of structures and the avoidance or mitigation of engineering disasters. In this paper, empirical mode decomposition and support vector machine model are used to study the prediction of wind speed and earthquake acceleration. The main contents are as follows: 1) the unstable characteristics of wind and earthquake and the methods used to predict non-stationary data series are analyzed. In this paper, the characteristics of traditional prediction methods are compared with the existing prediction methods. (2) the principles of support vector machine based on statistical theory and least-squares support vector machine learning method based on structural risk minimization are introduced. In order to improve the accuracy of model prediction, particle swarm optimization algorithm is introduced to establish the model kernel function and parameter selection. Concrete modeling steps. Finally, wind speed prediction and earthquake acceleration prediction are taken as examples to verify the effectiveness of this prediction method. 3) an empirical modal decomposition method combined with multistep prediction least squares support vector machine is proposed. The nonlinear time series analysis of wind speed is modeled and predicted. Firstly, the wind speed dynamic signal is decomposed by empirical mode, and the original signal is decomposed into several eigenmode functions of different characteristic scales (frequencies). The prediction model of least squares support vector machine (LS-SVM) based on multistep prediction is established. The stationary IMF components of different frequency bands are predicted and the final prediction values are obtained by equal-weight summation of the predicted values of each component. It is found that the prediction accuracy of the wind speed prediction method based on the combination of EMD and multistep LS-SVM is higher than that of the single LS-SVM forecasting method. On this basis, the PSO optimization algorithm is applied to optimize the selection of model parameters, and the optimized prediction result is obtained. (4) because seismic acceleration also has the characteristics of non-stationary time series, so, Similarly, the EMD-LS-SVM prediction method based on PSO optimization is used to model and predict the acceleration sequence. The calculation results verify the applicability of the combined method in the prediction of non-stationary time series. Finally, the results of this paper are analyzed and summarized, and the future research direction is prospected.
【学位授予单位】：河北农业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TV312
本文编号：2233190
[Abstract]:In water conservancy projects, there are many tall or long structures, such as high dams and their top buildings, working bridges and long span aqueducts. However, time series loads such as wind and earthquake have a great impact on these structures, and some projects are affected by accidents as a result. Or even disaster. Further study on the characteristics and rules of time series loads such as wind and earthquake, and prediction of them will be helpful to the vibration control of structures and the avoidance or mitigation of engineering disasters. In this paper, empirical mode decomposition and support vector machine model are used to study the prediction of wind speed and earthquake acceleration. The main contents are as follows: 1) the unstable characteristics of wind and earthquake and the methods used to predict non-stationary data series are analyzed. In this paper, the characteristics of traditional prediction methods are compared with the existing prediction methods. (2) the principles of support vector machine based on statistical theory and least-squares support vector machine learning method based on structural risk minimization are introduced. In order to improve the accuracy of model prediction, particle swarm optimization algorithm is introduced to establish the model kernel function and parameter selection. Concrete modeling steps. Finally, wind speed prediction and earthquake acceleration prediction are taken as examples to verify the effectiveness of this prediction method. 3) an empirical modal decomposition method combined with multistep prediction least squares support vector machine is proposed. The nonlinear time series analysis of wind speed is modeled and predicted. Firstly, the wind speed dynamic signal is decomposed by empirical mode, and the original signal is decomposed into several eigenmode functions of different characteristic scales (frequencies). The prediction model of least squares support vector machine (LS-SVM) based on multistep prediction is established. The stationary IMF components of different frequency bands are predicted and the final prediction values are obtained by equal-weight summation of the predicted values of each component. It is found that the prediction accuracy of the wind speed prediction method based on the combination of EMD and multistep LS-SVM is higher than that of the single LS-SVM forecasting method. On this basis, the PSO optimization algorithm is applied to optimize the selection of model parameters, and the optimized prediction result is obtained. (4) because seismic acceleration also has the characteristics of non-stationary time series, so, Similarly, the EMD-LS-SVM prediction method based on PSO optimization is used to model and predict the acceleration sequence. The calculation results verify the applicability of the combined method in the prediction of non-stationary time series. Finally, the results of this paper are analyzed and summarized, and the future research direction is prospected.
【学位授予单位】：河北农业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TV312
【共引文献】
中国博士学位论文全文数据库 前10条
1 王敏;分布式电源的概率建模及其对电力系统的影响[D];合肥工业大学;2010年
2 张颖媛;微网系统的运行优化与能量管理研究[D];合肥工业大学;2011年
3 李智;风电规律预测及对电网运行影响的研究[D];山东大学;2011年
4 刘辉;铁路沿线风信号智能预测算法研究[D];中南大学;2011年
5 刘伟;综放工作面煤矸界面识别理论与方法研究[D];中国矿业大学（北京）;2011年
6 杨志军;铁磁性平板腐蚀缺陷多通道漏磁信号的反演与重构[D];东北石油大学;2011年
7 陈建国;基于独立分量分析的机械故障特征提取及分类方法研究[D];大连理工大学;2011年
8 吴震宇;内燃机故障诊断若干理论与相关技术的研究[D];东北大学;2010年
9 高阳;风电场风电功率预测方法研究[D];沈阳农业大学;2011年
10 王扬;风电短期预测及其并网调度方法研究[D];浙江大学;2011年
，本文编号：2233190
本文链接：https://www.wllwen.com/kejilunwen/shuiwenshuili/2233190.html
