相关向量机理论在风电功率实时预测中的应用

发布时间：2018-05-29 22:02

本文选题：风电功率 + 超短期预测　；参考：《东北电力大学》2017年硕士论文

【摘要】：在我国,风电是目前最有潜力的,可以大力发展的非水电可再生能源。但同时风能的诸多自身特性,包括随机性、不确定性等,使大规模风电并网存在一些困难。为了实现大规模风能的开发利用,以超短期风电功率预测为背景,吉林省多个风电场的实测数据为基础,从风电功率的数据补齐、多步滚动的预测方法、风电功率预测的不确定性分析以及预测误差的非参数拟合四个方面进行了全面的分析与研究。对于风力发电的特性分析、功率预测、储能配置等研究都需要在历史数据的基础上进行展开,但实际中往往会由于各种原因导致数据不完整,缺失的数据可能会使系统变得混乱、难控制,或者存在越来越多的不确定性变化,这些情况都会对后续的分析估计造成很大的障碍。基于最大相关最小冗余原则对风电场风电功率数据进行补齐,首先分析得出与功率有关的变量,然后根据互信息理论,对变量通过最大相关最小冗余的原则进行特征选取,挖掘特征与功率之间的联系,最后根据这种联系对功率数据进行补齐。结果表明特征选取是对高维数据进行降维的有效办法,从原始特征集中选出特征子集,保留原始特征集的有效信息,从而补齐缺失的数据。风电功率预测的准确率越高,风能的利用率越高,因此,需要确定合理有效的预测方法,建立多步滚动的风电功率预测模型。相关向量机(RVM)是一种稀疏概率模型的学习机,具有很好的泛化学习能力,能有效地预测风电功率并且运行时间极快。同时引入集合经验模态分解(EEMD),将功率数据的初始序列分解成若干组平稳的序列,该方法可以显著提高预测精度,缩短运行时间。由于任何预测都具有不确定性,因此带有置信区间的单点预测范围可以降低电网和风电场运行的风险,整个系统的运行也就更安全稳定。对风电功率预测的不确定性进行分析,可以把预测功率的单一值转化成功率的估计区间。结果表明相关向量机的预测模型可以提供给定置信水平下的预测波动范围。对预测误差进行拟合分布评价,通过对预测误差的分布特征可以分析得出非参数估计与预测方法、预测时间间隔、预测误差概率分布形态以及风电场装机容量的关系,从而使系统稳定持续地运行。结果表明非参数估计分布模型对不同规模的风电场和不同条件的分布均能较好地拟合,其中单峰的拟合效果更好。
[Abstract]:Wind power is the most potential non-hydropower renewable energy in China. But at the same time, wind energy has many characteristics, such as randomness and uncertainty, which makes large-scale wind power grid difficult. In order to realize the development and utilization of large-scale wind energy, based on the forecast of ultra-short-term wind power and the measured data of several wind farms in Jilin Province, the prediction method of wind power compensation and multi-step rolling is introduced. The uncertainty analysis of wind power prediction and the nonparametric fitting of prediction error are analyzed and studied comprehensively. For wind power generation characteristics analysis, power prediction, energy storage configuration and other studies need to be carried out on the basis of historical data, but in practice, due to various reasons, the data are often incomplete. The missing data may make the system chaotic, difficult to control, or there are more and more uncertain changes, which will cause great obstacles to the subsequent analysis and estimation. Based on the principle of maximum correlation and minimum redundancy, the wind power data of wind farm is compensated. Firstly, the variables related to power are analyzed, and then, according to the mutual information theory, the variables are selected by the principle of maximum correlation and minimum redundancy. The relation between feature and power is mined, and the power data is corrected according to this relation. The results show that feature selection is an effective method to reduce the dimension of high-dimensional data. The feature subset is selected from the original feature set, and the effective information of the original feature set is retained. The higher the accuracy of wind power prediction, the higher the utilization rate of wind energy. Therefore, it is necessary to determine a reasonable and effective forecasting method and to establish a multi-step rolling wind power prediction model. Correlation vector machine (RVM) is a kind of learning machine with sparse probability model. It has good generalization ability and can effectively predict wind power and run very fast. At the same time, the set empirical mode decomposition (EMD) is introduced to decompose the initial sequence of power data into a number of stationary sequences. This method can significantly improve the prediction accuracy and shorten the running time. Because of the uncertainty of any prediction, the single point prediction range with confidence interval can reduce the risk of power grid and wind farm operation, and the operation of the whole system is safer and more stable. By analyzing the uncertainty of wind power prediction, the single value of predicted power can be transformed into the estimated interval of success rate. The results show that the prediction model of correlation vector machine can provide the range of predicted fluctuations at a given confidence level. The relationship between nonparametric estimation and prediction method, prediction time interval, probability distribution pattern of prediction error and installed capacity of wind farm can be obtained by analyzing the distribution characteristics of prediction error. Thus, the system runs steadily and continuously. The results show that the non-parametric distribution model can fit the distribution of wind farms of different scale and different conditions, and the fitting effect of single peak is better.
【学位授予单位】：东北电力大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM614

【参考文献】