基于SVR异常值检测和VMD分解的GRNN和Markov联合模型在风速预测中的应用

发布时间：2018-04-15 16:14

本文选题：异常值检测 + 支持向量机回归　；参考：《兰州大学》2017年硕士论文

【摘要】：建立模型的优劣基于所搜集数据的质量。由于各种不确定因素的影响,数据一般都存在异常值,所以对数据进行异常值检测是十分必要的。此外,对时间序列数据进行分解也是比较常用的一种数据预处理方法,分解后的子序列能够反映出数据的不同特征,可为数据的进一步分析提供准确的信息。本文以风速作为研究对象。一方面,风力资源是清洁能源,另一方面,风力资源为风能发电提供了充足的来源,属于可持续能源,所以对风速的预测就显得十分重要。本文基于西班牙Sotavento Galicia风场的风速数据,首先运用支持向量机回归(Support Vector Regression, SVR)方法对原始数据进行异常值检测。然后运用变分模态分解(Variational Mode Decomposition, VMD)方法,对剔除异常值后的序列进行分解。接下来运用广义回归神经网络(Generalized Regression Neural Network, GRNN)对分解后的子序列分别进行预测。最后将所有子序列的预测值进行相加并运用马尔科夫(Markov)过程进行误差修正得到最终的预测值。实验结果表明,SVR异常值检测方法、VMD分解方法和Markov误差修正模型对于模型预测精度的提高是有用的。此外,本文将提出的模型与其它常规方法进行了对比,证明了其效率和估计性能。
[Abstract]:The establishment of the model is based on the quality of the collected data.Because of the influence of various uncertain factors, there are outliers in the data, so it is necessary to detect the outliers.In addition, decomposing time series data is also a commonly used method of data preprocessing. The decomposed sub-sequences can reflect the different characteristics of the data and provide accurate information for the further analysis of the data.This paper takes the wind speed as the research object.On the one hand, wind energy is a clean energy, on the other hand, wind resources provide sufficient sources for wind power generation, so it is very important to predict wind speed.In this paper, based on the wind speed data of Sotavento Galicia wind field in Spain, the support vector machine regression support Vector method is used to detect the outliers of the original data.Then the variational mode decomposition (VMD) method is used to decompose the sequence after eliminating the outliers.Then generalized Regression Neural network (GRNN) is used to predict the decomposed subsequences.Finally, the prediction values of all sub-sequences are added together and the final prediction value is obtained by using Markov Markov process to correct the error.The experimental results show that the VMD decomposition method and the Markov error correction model are useful for improving the prediction accuracy of the model.In addition, the proposed model is compared with other conventional methods, and its efficiency and estimation performance are proved.
【学位授予单位】：兰州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM614;P412.16

【参考文献】