风电场超短期风电功率多步预测及可预测性研究

发布时间：2018-05-03 17:13

本文选题：风电功率 + 波动特性　；参考：《东北电力大学》2017年硕士论文

【摘要】：随着全球化石能源的消耗和电力需求的增长,开发可再生能源进行电力生产已成为全球各国关注的焦点,风能相比于其它能源具有显著优势,成为近年来发展最为迅速的新能源发电技术。然而受自然界风的影响,风电具有随机性、间歇性和不确定性,大规模风电并网对电力系统安全稳定运行带来了不利影响,因此,对风电功率进行预测具有重要的现实意义。本文以实际风电场为研究对象,对风电功率波动特性、超短期风电功率预测方法、风电功率预测误差和风电功率可预测性进行了相关研究。针对大规模风电场输出功率波动的时空分布特性,提出基于混合分布模型的风电功率波动特性概率分布描述方法,分别对相同机组数量不同采样间隔和相同采样间隔不同机组数量的风电功率波动的时空分布进行描述,在此基础上分析了风电功率波动变化率的累积概率随采样间隔和机组数量的变化规律,对风电功率预测具有重要意义。基于相关性分析和K近邻算法,提出一种多输出模型的风电功率超短期预测方法,实现了预测精度的提高。同时,为增强预测方法的信息互补性,组合单一预测方法,建立基于自适应神经模糊推理系统的组合预测模型,实现对单一风电功率预测结果的优化。以风电场的实测数据为例,进行仿真分析,验证了两种预测模型的有效性,并分析汇聚效应对预测结果的影响,有利于风电功率的预测精度的提高。针对不同功率水平的预测误差所呈现不同的分布特性,提出基于混合分布模型来描述风电功率预测误差的分布特性,通过与其它分布模型的对比,验证了有效性。由于风电功率数据存在显著的时间相依结构,对预测功率水平不同进行划分,以划分区段内的预测误差为统计样本,进行预测误差分布的概率密度拟合,进而求解相应的累积概率。风电功率预测误差的概率分析可为风电功率预测精度的提高和不确定性分析提供依据。针对风电功率预测方法无法实现完全绝对地无差预测的客观事实,提出了风电场风电功率时间序列的可预测性概念,并利用近似熵和可预测系数对风电功率时间序列的可预测性分别进行定量分析,在此基础上分析了不同机组数量汇聚时其近似熵的变化规律,验证了所提指标的有效性。风电功率可预测性研究可在同一平台上客观地评价预测方法的优劣,还可以为不同风电场确定切实可行的预测精度考核指标提供依据。
[Abstract]:With the global consumption of fossil energy and the increase of electricity demand, the development of renewable energy for electricity production has become the focus of attention in the world. Wind energy has a significant advantage over other energy sources. In recent years, the most rapid development of new energy generation technology. However, due to the influence of natural wind, wind power has randomness, intermittency and uncertainty. Large-scale wind power grid connection brings adverse effects on the safe and stable operation of power system, so it is of great practical significance to predict wind power. In this paper, the characteristics of wind power fluctuation, the prediction method of ultra-short-term wind power, the prediction error of wind power and the predictability of wind power are studied. According to the temporal and spatial distribution characteristics of large scale wind farm output power fluctuation, a method of describing the probability distribution of wind power fluctuation characteristics based on mixed distribution model is proposed. The time-space distribution of wind power fluctuation with the same number of units with different sampling intervals and the same sampling interval with different number of units is described respectively. On this basis, the cumulative probability of wind power fluctuation rate with sampling interval and the number of units is analyzed, which is of great significance for wind power prediction. Based on correlation analysis and K-nearest neighbor algorithm, an ultra-short-term wind power prediction method based on multi-output model is proposed, and the prediction accuracy is improved. At the same time, in order to enhance the information complementarity of forecasting methods, a combined prediction model based on adaptive neural fuzzy inference system is established to optimize the prediction results of single wind power. Taking the measured data of wind farm as an example, the validity of the two prediction models is verified, and the influence of convergent effect on the prediction results is analyzed, which is beneficial to the improvement of the prediction accuracy of wind power. In view of the different distribution characteristics of prediction errors at different power levels, a hybrid distribution model is proposed to describe the distribution characteristics of wind power prediction errors. The validity of the proposed model is verified by comparison with other distribution models. Because of the significant time-dependent structure of wind power data, the predicted power level is divided into different levels, and the prediction error in the division section is taken as the statistical sample to fit the probability density of the prediction error distribution. Then the corresponding cumulative probability is solved. The probabilistic analysis of wind power prediction error can provide the basis for the improvement of wind power prediction accuracy and uncertainty analysis. In view of the objective fact that wind power prediction method can not achieve absolute absolute difference prediction, the concept of predictability of wind power time series in wind farm is put forward. The approximate entropy and the predictable coefficient are used to quantitatively analyze the predictability of wind power time series respectively. On this basis, the variation law of approximate entropy of different units is analyzed, and the validity of the proposed index is verified. The predictive study of wind power can objectively evaluate the merits and demerits of forecasting methods on the same platform, and can also provide the basis for determining feasible evaluation indexes of prediction accuracy for different wind farms.
【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】