【作者】 刘兴杰；

【导师】 杨奇逊； 米增强；

【作者基本信息】 华北电力大学 ， 电力系统及其自动化， 2011， 博士

【摘要】 风电具有波动性和间歇性,大规模风电接入系统对电网的安全稳定运行带来了严峻挑战。风电输出功率预测是解决大规模风电接入问题的关键技术之一,可以为风、光、储、水等多种电源多点接入互补运行和接入电网的调度控制提供支持。目前,国内针对风电输出功率预测方法的研究还不够深入,预测系统的开发相对较少,缺乏成熟的实践经验。可见,对风电输出功率预测方法进行研究具有重要的理论意义和现实意义。本文以两个典型的风电场为对象,对风速预测方法、功率曲线建模方法和风电输出功率预测方法展开了系统的研究,以探求精度更高的风电输出功率预测方法,并为风电输出功率预测技术路线的制定和预测系统的开发提供指导。同时,根据需求分析研发了风电输出功率预测系统。主要成果如下:1、风速预测是风电输出功率预测的重要基础,对风速预测方法进行了研究。1)提出了一种以经验模式分解(Empirical Mode Decomposition,EMD)为核心的风速预测新方法,通过与常规风速预测方法对比分析,对基于EMD的风速单步和多步预测方法进行了系统研究。结果表明,由于经过EMD处理,原始风速时间序列被自适应地分解为多个相互正交的分量,从而降低了建模和预测难度,提高了风速预测精度。同时,基于EMD的风速直接多步预测模型中,对重构所得分量多步预测结果进行自适应叠加,能够削减偏差较大分量对整体预测结果的影响,进一步提高多步预测精度。2)提出了一种基于混沌分析和RBF的风速直接多步预测方法,通过与其它常规方法对比分析,表明由于对原始风速时间序列进行了相空间重构,能够更好地反映风速内在变化规律,并且为RBF网络结构的确定和训练样本的选择提供依据,故其直接多步预测精度较RBF更高。2、基于功率曲线进行风电功率预测是一种有效方法,对功率曲线建模方法进行了研究,提出了一种新方法-指数遗忘建模法,通过测试表明,以实测风速或预测风速作为输入求取功率,指数遗忘法均具有较高的精度。该方法物理意义明确,简单易行,适用于风电输出功率预测。3、对风电输出功率预测方法进行了系统研究,包括两种预测途径和多种预测算法。结果表明,进行风电输出功率超短期预测时,应根据不同的预测对象选择相应预测方法,而预测时采用基于EMD的算法能够普遍提高预测精度。在此基础上,提出了适合我国国情的风电输出功率预测技术路线。同时指出,当进行风电输出功率短期预测时,在数值天气预报数据积累不足和不够成熟稳定的情况下,采用功率曲线法较神经网络法具有更高的精度。4、开发了风电输出功率预测系统,实现了对风电输出功率的超短期和短期预测。在多个风电场和地区电网调度的实际运行结果表明,系统安全可靠、用户界面友好,可操作性强,能够高精度地实现预测功能,且不受风电机组检修、停机、风电场扩建的限制。更多还原

【Abstract】 The intermittent and volatility of wind energy brings rigorous challenges for the safety and stability of electric power system, when large scale wind access system. An effective technique to solve this proble is to predict wind power. Also, the wind power prediction can provide supprots for the complementary opperation of wind power, photovoltaic power, storage energy, hydropower and the diapatching control of power system. The prediction approaches of wind power, however, are not be researched in depth enough in China. Moreover, the corresponding prediction system is less. So it is significant theoretically and practically to research the prediction approach of the wind power.To explore the prediction approach of wind power with higher accuracy, the wind speed prediction, the power curve model and the power prediction have been systematically studied in this paper based on two typical wind farms. The research achievements were expected to offer guidance for determining the prediction technique route and developing the prediction system.The main achievements are as follows:The wind speed prediction approach was deeply studied firstly, due to that it can provide support for the wind power prediction. 1) A new prediction approach for wind speed based on EMD was proposed. The wind speed prediction results were compared with those gotten through the other conventional methods. The results show that when applying the introduced approach, both the wind speed single-step prediction accuracy and the multi-step one get improved. The reason lies in that the original wind speed series are self-adaptively decomposed into some normal components after the preprocessing of EMD. As a result, the coupling and interferences among the components are weakened. So the prediction models are easier to build, furthermore, the prediction accuracy is enhanced. When the multi-step prediction for wind speed was performed based EMD, the prediction results of the reconstitution components were overlapped adaptively. In doing this, the influences of the component with the more drifts on the whole prediction effects are weakened, so the prediction precision can be further improved. 2) A novel direct multi-step prediction approach for wind speed based on chaos analysis was presented. According to the phase space reconstruction theory, the phase space of original wind speed is reconstructed, which could reflect the inner rules embedded in wind speed and is helpful for the determination of RBF network. As a result, the prediction accuracy gets increased compared to the RBF method.The power curve plays an important role in the power prediction. So the modeling of power curve was studied. A power curve modeling approach was introduced, which was nominated as“exponentially forgetting modeling”. The simulation results show that its precision is satisfied, no matter the input is real wind speed and predicted wind speed. Compared to RBF, the introduced model is more suitable to be applied in power prediction due to it is easy to employ and possesses explicit physical meaning.Based on the above investigations, the power prediction approach was systematically researched, which mainly focused on the two prediction ways and some algorithms. The application example shows when the ultra-short power is predicted, the different prediction scheme should be chose according to the predicted objects. But no matter which scheme is confirmed, the algorithm based on EMD has highest accuracy. On the basis of these, the technology options of power prediction suited for China is proposed. While the short power is predicted, the power curve approach has better effects comparing with RBF, on the premise of the NWP is not matured.The power prediction system was developed in this paper, which had been installed in some wind farms and the electric dispatching centers. The practices indicate that the ultra-short and short power prediction with higher accuracy can be real-timely implemented by means of the system. At the same time, the system is safe, user-friendly and easy to operate.更多还原