基于数值天气预报信息融合的并网风电场短期发电功率预测研究

Numerical weather prediction and forecasting methods are important factors to affect the accuracy of wind power forecasting. This work makes a study on the forecasting model of wind power generation from the angle of how to use various NWP data, improves the modeling method, and increases the precision of wind power forecasting. It firstly analyzes NWP information of multiple grid points and locations from a single weather prediction system and uses the rough set theory and the principal component analysis to explore more meteorological factors to impact wind power generation; secondly uses the cluster analysis to classify the NWP samples and establishes an integrated forecasting model which corresponds to different weather situation; thirdly examines the relationship between the actual wind speed and the outputs of NWP models from multiple weather prediction systems and uses the local lazy learning (LL) algorithm to construct the wind speed forecasting model based on adaptive data fusion of multiple models, and uses the nonparametric regression method to realize the transformation from wind speed to wind power; fourthly assesses the uncertainty of wind power forecasting, uses the lognormal distribution to build the distribution function of wind power forecasting error, and realizes the interval forecasting; and finally develops a set of software for wind power forecasting based on various weather prediction and provides the basis for the management and operation of wind farms.

数值天气预报（NWP）和预测方法是影响风力发电功率预测准确性的重要因素。课题从如何利用多种数值天气预报数据的角度来研究风力发电功率预测模型，改善建模方法，实现风电功率预测精度的提高。首先，单一天气预报系统下，解析风电场周围多网格点多高度NWP信息，采用粗糙集理论和主成分分析法探索出影响风电场发电功率的更多的气象因素；其次，利用聚类分析法对NWP样本分类，建立集成预测模型，各种天气状态下有相对应的模型；再次，多个天气预报系统下，研究多个NWP模型输出与实际风速间的关系，利用局部懒惰算法建立基于多模型自适应数据融合的风速预测方法，利用非参数回归法建立风速-功率转换模型；然后，对风电功率预测不确定性进行评估，利用对数正态分布建立风电功率预测误差的分布函数，实现风电功率概率区间预测；最后，开发一套基于多种气象信息的适用于大型风电场发电功率预测的应用软件，为风电场的管理和运行提供依据。

本项目以数值天气预报为基础，以提高风电功率预测系统的精度为目标，分别从单位置天气预报和多位置天气预报出发，结合一定的研究方法来完成风电场发电功率的建模和预测。在数据处理与分析方面，提出了一种互补集合经验模态分解方法，互补集合经验模态分解可以降低风电功率时间序列的非平稳性，混沌理论可以提取数据的特征；在基于单位置数值天气预报建模方面，利用多种聚类方法划分天气状态，提出了一种基于数学形态学聚类的方法提取相似的发电功率，利用多种优化方法对模型参数进行优化，提出了一种果蝇优化算法实现模型参数的最优化；在基于多位置数值天气预报建模方面，提出主成分分析和奇异值分解的方法对多种数值天气预报数据进行融合处理，建模时提出了一种卡尔曼滤波算法对数值天气预报进行修正并与机器学习相结合的模型，可以大大提高预测精度。本项目基于数值天气预报对风电功率预测模型及预测方法进行研究，在天气预报数据精度不高或者存在系统误差的情况下，能一定程度地改善风电功率预测的精度，为风电更好地参与电网调度提供依据。