【作者】 王富强；

【导师】 韩璞； 王东风；

【作者基本信息】 华北电力大学 ， 控制理论与控制工程， 2013， 博士

【摘要】 随着全球能源问题的日益严峻,风能作为一种重要的可再生能源,其装机容量和单机容量迅速提高,然而风电本身的波动性给并网后的电力系统带来不良冲击,影响电力系统的安全平稳运行。为了降低风电对电网的冲击,合理调度风能资源,对风电场风能进行预测是十分重要的。关于中长期的风速预测,广大学者已进行了广泛的研究,并且取得了不错的效果；而超短期和短期的风速具有很强的随机性和非平稳性,其预测效果不是很理想。本文围绕风速时间序列随机性和非平稳特性的几个关键技术问题,展开了以下研究：(1)多步预测策略选择的研究。在预测策略层面上,对短期风速预测进行研究,分析多种传统预测策略的特点,针对其在短期风速预测中的局限性,提出了预测误差补偿策略,并将其与直接多输出策略结合,得到了补偿-直接多输出策略,有效地提高了短期风速多步预测精度。(2)短期风速时间序列趋势项的提取。详细介绍了小波分解和经验模态分解的基本原理,说明为了提高多步预测的预测精度,对时间序列进行趋势项提取是十分有效的。根据短期风速的特点,重点介绍了小波分解和经验模态分解趋势项的提取方法,研究了小波高频/低频分量预测、部分高频/低频分量预测和低频分量预测的三种方法的特点；分析了小波分解与经验模态分解在短期风速预测中的效果,得出了经验模态分解理论更加适用于短期风速时间序列的趋势项提取问题的结论。(3)短期风速时间序列的混沌特性以及相空间重构。由于短期风速特性具有很强的随机性和非平稳性,首先利用混沌理论分析短期风速时间序列具有混沌特性,在此基础上进行相空间重构,确定嵌入维m和延迟时间τ,从而确定预测模型输入向量,进行短期风速预测,显著提高了预测精度。(4)基于组合预测权值的短期风速组合预测。提出了采用组合理论解决BP神经网络的隐节点难于确定的问题和相空间重构中嵌入维计算结果不一致的问题；研究了线性组合方法和非线性组合方法；并且将其与经验模态分解理论结合使用,显著提高了预测精度。(5)基于最优预测模型的短期风速组合预测。提出了一种基于多属性决策理论的多步风速预测模型方法。该方法从“成本”的角度,综合考虑多个性能指标,其中包括历史数据的预测性能分析和未来预测值信息在内的属性,确定某个单项预测模型的预测值作为组合预测的预测结果,从而提高预测精度。(6)风电场风速预测专家系统。针对不同场址的不同时间段的风速时间序列特性不同的特点,结合本论文所研究的预测模型以及专家系统理论,建立了以组合预测理论为框架的风电场风速预测专家系统。(7)基于几何布朗运动的短期风速数值模拟。该方法运用伊藤定理模拟几何布朗运动,计算模型简单,应用方便,通过模拟数值的混沌特性检验以及功率谱的检验,说明模拟数值符合短期风速时间序列的基本特征。更多还原

【Abstract】 As global energy problem becomes increasingly serious, wind power is a kind of important renewable energy. The installed capacity and single capacity increase quickly. The volatility of wind power influences the safety and stability of power grid. In order to reduce the impact of wind power to power grid and have the rea-sonable scheduling wind energy, wind power prediction is very important. Involved in wind speed prediction about medium and long term, the scholars have made ex-tensive research, and achieved good effect. But super short term and short-term wind speed has strong randomness and instability, and its prediction effect is not very ideal.The following studies are carried out around some key technical issues of randomness and non-stationary characteristics for wind speed time series:(1) The choice of multi-step prediction strategies. In the level of the prediction strategy, studying short-term wind speed prediction and analyzing the characteris-tics of a variety of traditional prediction strategies have been done. According to the limitations of traditional prediction strategies for short-term wind speed predic-tion, error compensation strategy is proposed. It is combined with DirMO strategy to get ComDirMO strategy. ComDirMO strategy effectively improves the mul-ti-step prediction accuracy.(2) The trend term extraction of time series for short-term wind speed. The basic principles of wavelet decomposition and empirical mode decomposition are introduced in detail. In order to improve multi-step prediction accuracy, trend term extraction is very effective. According to the characteristics of short-term wind speed, focusing on the trend term extraction method of wavelet decomposition and empirical mode decomposition, it studies high-frequency and low-frequency com-ponent prediction, part-high-frequency and low-frequency component prediction and low-frequency component prediction. Wavelet decomposition and empirical mode decomposition are analyzed in the short-term wind speed prediction. It is concluded that empirical mode decomposition theory is more suitable for the trend term extraction of short-term wind speed time series.(3) The chaotic characteristic and phase space reconstruction for short-term wind speed time series. Due to the strong randomness and instability characteristic of the short-term wind speed, it has the chaos characteristic by chaos theory. Based on the phase space reconstruction, embedding dimension m and delay time τ are determined to have the input vector. It can improve the prediction accuracy. (4) The combination prediction based on the combination of weights for short-term wind speed. The combination theory is put forward to solve the problem of BP neural network’s hidden node, and the inconsistent problem of embedding dimension results in phase space reconstruction. The linear combination method and the nonlinear combination method are studied. Combined with the empirical mode decomposition theory, it improves the prediction accuracy.(5) The combination prediction based on the optimal prediction model for short-term wind speed. Based on the theory of multi-step wind speed prediction, a multiple attribute decision making model is put forward. In the view of "cost", multiple performance indicators are considered, which include the prediction per-formance analysis of the historical data and the attributes of the future prediction information. The predictive value of a prediction model is determined as the com-bination prediction result to improve the prediction accuracy.(6) Expert system of wind speed prediction. Wind speed time series of differ-ent times for different sites have different characteristics. Combination prediction model and expert system theory are studied in this paper, the expert system of wind speed prediction is established, which takes the combination prediction theory as the framework.(7) The numerical simulation based on geometric Brownian motion for short-term wind speed. This method uses ITO theorem to simulate geometric Brownian motion. The calculation model is simple and easy application. By the chaotic characteristic and the power spectrum test, it shows that the numerical sim-ulation is in accord with the basic characteristics of short-term wind speed time se-更多还原