基于历史气象数据的风电场风速和风功率预测研究

发布时间：2018-01-05 20:10

本文关键词：基于历史气象数据的风电场风速和风功率预测研究　出处：《东北电力大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着化石能源的大量消耗和环保压力的与日俱增,风能作为可再生、清洁能源受到世界各国越来越高的重视。受地形、气候及周围环境等众多因素的影响,风力发电具有强烈的间歇性、波动性和不确定性,给电力系统的发电规划和经济调度带来了危害,不利于传统电网的安全、稳定运行,阻碍了风电进一步的推广和应用。因此,大规模的风电并网运行需要精准的风电功率预测。准确的风速是风功率预测的关键因素,因此准确的风速预测也具有重要的现实意义。风速时间序列易受温度、气压、湿度等气象因素的影响,且具有很高的输入维度和较强的非线性,故不易精准预测。极限学习机(ELM)及其优化模型结构简单、运算效率高、泛化能力强,可以根据问题需要灵活地选择隐含层节点数和激活函数类型,适用于复杂的非线性风速预测工作。本文主要针对风速和风功率短期预测进行研究,相关内容如下:针对风速属性多、难以精度预测等特点,本文采用互信息对属性序列与风速、功率序列之间的相关性进行分析,并采用最大相关最小冗余(mRMR)进行属性选择以降低输入维度。之后,采用皮尔逊相关性系数(PCC)对预测输入属性数据进行加权处理,以便凸显关联程度高属性的重要度,提高预测精度。最后,采用ELM及其优化模型开展预测研究。为了降低风速序列信号的波动性和非线性对电力系统的影响,进一步提高风速预测精度,可以采用信号分解方法对其处理以得到相对稳定的子序列。本文采用变分模态分解(VMD)对初始时间序列进行分解处理,得到一系列具有一定周期性、规律性且相对稳定的子序列。针对每一个子序列,运用偏自相关函数(PACF)筛选出关联程度高的元素集合,确定网络模型的输入,选用泛化能力强的加权正则化极限学习机(WRELM)构建多步滚动预测模型,开展短期风速预测。参照风电机组的风速-功率曲线,根据预测风速可直接得出对应时刻的风功率。针对风电功率大量的相关属性,本文参照风速预测方法,在互信息相关性分析的基础上依据m RMR对候选属性集合进行属性选择,并采用PCC对其进行重要度加权,采用优化ELM网络进行预测拟合。最后,基于MATLAB 8.5(2015a)软件平台本文采用风电场实测数据对上述内容进行仿真实验,结果验证新方法的准确性和实用性。
[Abstract]:With the grow with each passing day massive consumption of fossil fuels and environmental pressure, wind energy as a renewable and clean energy has attracted more and more attention all over the world. The terrain, climate and environment and many other factors, wind power has strong intermittency, volatility and uncertainty, which brings great harm to the economic power generation planning and scheduling system that is not conducive to the traditional power grid safety, stable operation, hindered the popularization and application of wind power further. Therefore, large-scale wind power grid connected wind power requires accurate forecasting. Accurate wind speed is the key factor for wind power prediction, so it has important practical significance to speed accurate prediction of wind speed time. The sequence is easily affected by the temperature, pressure, humidity and other meteorological factors influence, and has very high input dimension and strong nonlinearity, so it is not easy to accurately predict. The extreme learning machine (ELM) and The model has the advantages of simple structure, high operation efficiency, strong generalization ability, can choose the number of hidden layer nodes and activating function according to the need, applicable to complex nonlinear wind speed prediction. This paper focuses on wind speed and wind power forecasting, relevant content are as follows: according to the wind speed prediction accuracy is difficult to attribute. The characteristics of mutual information of attribute sequence and the correlation between wind speed, power series is analyzed, and the optimization (mRMR) for feature selection in order to reduce the input dimension. Then, by using Pearson correlation coefficient (PCC) were weighted attribute data to predict the input, in order to highlight the importance of a high degree of correlation attributes and improve the prediction accuracy. Finally, carry out the prediction research using ELM and its optimization model. In order to reduce the wind speed sequence signal and nonlinear wave of electricity The influence of the system, further improve the prediction accuracy of wind speed, can be used for the signal decomposition method to obtain relatively stable sequence. Using variational mode decomposition (VMD) to the initial time series decomposition, obtained a series of highly periodic sequence regularity and relatively stable for each. Sub sequences, using partial autocorrelation function (PACF) selected set is associated with a high degree of network elements, determine the model input, using weighted regularized extreme learning machine strong generalization ability (WRELM) to construct multi-step prediction model, carry out short-term wind speed forecast. Wind speed reference power curve of wind turbine, according to the forecast the wind speed and wind power can be directly obtained. The corresponding time for wind power is related to a large number of attributes, according to the wind speed prediction method based on correlation analysis, mutual information based on M RMR to the candidate Attribute set for attribute selection, and PCC is used for its important degree, is predicted by the fitting optimization ELM network. Finally, based on the MATLAB 8.5 (2015a) software platform used in this simulation experiment on the content of the measured data of a wind farm, the results verify the accuracy and practicability of the new method.

【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】