【作者】 韩爽；

【导师】 杨勇平； 刘永前；

【作者基本信息】 华北电力大学（北京） ， 热能工程， 2008， 博士

【摘要】 风力发电进入了规模化发展阶段。但是风电具有间歇性和随机性的固有缺点,随着大量的风力发电接入电网,势必会对电力系统的安全、稳定运行以及保证电能质量带来严峻挑战,从而限制风力发电的发展规模。风电场功率短期预测是解决该问题的有效途径之一。中国的风电场大都是集中的、大容量的(百万千瓦级甚至千万千瓦级)风电场,而电网建设相对比较薄弱,因此,中国更需要进行风电场功率短期预测的研究,而目前中国在此领域的研究还处在起步阶段。在此背景下,选择风电场功率短期预测方法作为论文研究内容,对多种预测方法的应用进行了研究和改进,主要工作包括以下几个方面:首先对风电场测量参数和运行参数进行了预处理及其统计规律的研究,得到一些后面研究中要用到的结论。根据历史记录数据,使用随机时间序列分析方法针对一年12个月分别建立预测模型,进行提前1-6小时的风电场风速及功率预测;预测精度满足要求,并且发现预测误差随季节发生变化的规律。提出了基于平均训练误差的集成神经网络预测模型,用于1-6小时的风电场风速及功率预测,该模型具有明确的物理意义,可以显著提高网络的泛化能力;认为径向基神经网络比误差反向传播网络更适合风电场功率预测;与随机时间序列分析方法预测结果比较,神经网络方法可以提高预测精度,预测误差也呈现出随季节发生变化的规律。风电场功率曲线建模是功率预测的一个重要步骤,在论文中提出了四种风电场功率曲线的建模方法并进行比较;在此基础上提出了适合于中国大规模风电场的功率预测技术路线。为进一步提高预测精度,结合风电场功率随机性大的特点,并考虑风电场功率高阶矩的特征,建立了基于最大信息熵的风电场功率组合预测模型,应用实例表明,该方法可以有效提高预测精度。预测结果的不确定性信息非常重要,在论文中使用一种基于独立分量分析的条件概率方法,建立了风电场功率预测结果不确定性分析模型,该方法具有普适性,对使用数值气象预报与不使用数值气象预报的预测方法均适用。论文的研究基于两个风电场的实际运行和测量数据,结果具有实际意义和应用价值。更多还原

【Abstract】 Wind power has entered a rapid progress stage. But wind power has the disadvantages of intermittence and randomicity, which will bring challenge to the safety and stabilization of power grid and then restrict the scale of wind power development. Short-term wind power prediction is an effective approach for the above problem. The wind farms in China are mostly centralized and large scaled (mega kW or even kilo kW) ones, while the power grids construction is weak. Short-term wind power prediction is more needed in China. But the research in this field is at the beginning.This paper studied short-term wind power prediction methods and made some improvement. The main works are as follows:The study on statistical rules of wind farm parameters was done first. Some conclusions will be used later.Based on historical data, time series method was used to build prediction models for 12 months of one calendar year for wind speed and wind power prediction. The prediction precision met the engineering demand and the prediction errors varied with seasons.A kind of integrated ANN prediction model based on mean prediction error was proposed, which had clear physical meanings and can improve the generalization ability and prediction precision. The RBF network was considered to be more suitable for wind power prediction than BP network. Compared to the prediction results of time series method, the prediction results of ANN method are of higher precision. The prediction errors varied with seasons.Power curve modeling of wind farm is an important step in wind power prediction. Four kinds of different power curve modeling methods were put forward and compared. The wind power prediction route suitable for Chinese wind farm was put forward.According to the high randomicity of wind power, the wind power combined prediction model based on maximum information entropy theory was built to improve prediction precision, which considered the high order moments. The application example showed that the model can improve the prediction precision effectively.The uncertainty information of prediction result is very important. A method of conditional probability based on Independent Component Analysis was used here to build the uncertainty assessment model for wind power prediction. The model can be used for both prediction methods with Numerical Weather Predcition and without Numerical Weather Prediction.The study of this paper was based on the real data of two wind farms, so the conclusions have actual meanings and applied cost.更多还原