风电场风功率预测及最大风能追踪
发布时间：2018-12-21 10:45
【摘要】：随着能源的日益减少、环境污染越来越严重,尤其是雾霾正逐步覆盖着我国大部分城市,使用清洁无污染的能源便成为解决这些问题的一个重要选择。风能以其可再生、绿色、清洁环保等特性使得风力发电量在世界发电总量中所占比重正在不断加大。但由于风力发电的间歇性及不可控等特性使得电网接纳风电的能力受到抑制,风力发电的发展速度较为缓慢。对风功率进行较高精度的预测,可以有效地降低风电场输出波动对电力系统的不利影响,并且提高风力发电在电力市场中的竞争实力。同时,风电机组向更大容量、更高效率的趋势发展,就要求提高对风能的利用率,使得研究最大风能追踪具有重要意义。 本文针对风功率预测和最大风能追踪这两大风力发电中的实际问题,进行了深入研究。针对风功率预测,由于风功率与风速有着直接的函数关系,因此可以通过预测风速得到风功率的预测结果。对于风速序列的非线性及不平稳性,本文应用小波分解来降低。风速预测大多应用神经网络理论,目前还没有统一的神经网络输入量的选取方法,本文采用时间序列建模选择输入量的方法为神经网络选择输入量。综上所述,本文提出一种综合了小波理论、时间序列及神经网络的预测方法,即小波时序神经网络预测方法。该方法首先将原始风速进行小波分解为一个低频趋势信号和几个高频随机信号,然后对高频信号采用时间序列的方法建模,继而对小波分解后的两种信号利用BP神经网络进行建模：低频信号采用常规BP神经网络(输入量为最近的6个历史值),高频信号应用时间神经网络(输入量应用时间序列建模选择),最后将各信号的预测结果通过叠加进行重构,得到原始风速序列的最终预测结果。将风速预测结果输入到拟合的风机功率曲线,便可预测出风机功率。基于我国南方某风电场的实测数据进行验证,本文提出的预测方法具有较好的预测精度。针对最大风能追踪问题,本文在分析风力机运行特性的基础上,研究变速恒频双馈风力发电机组实现最大风能追踪的控制方法。为简化模型难度,对双馈异步发电机进行坐标变换,并采用定子磁链定向的矢量变换技术。简化后的模型可完成发电机的有功功率和无功功率的解耦控制,并且实现最大风能追踪的目标。应用Matlab/Simulink进行完整的风力发电系统的仿真研究,结果证明了所建模型的正确性。
[Abstract]:With the decrease of energy sources, environmental pollution is becoming more and more serious, especially haze is gradually covering most cities in China. The use of clean and non-polluting energy has become an important choice to solve these problems. Because of its renewable, green, clean and environmental characteristics, wind power generation in the world is increasing. However, due to the intermittent and uncontrollable characteristics of wind power generation, the ability of wind power grid to accept wind power is restrained, and the development of wind power generation is slow. The prediction of wind power with high accuracy can effectively reduce the adverse effect of wind farm output fluctuation on power system and improve the competitive power of wind power generation in power market. At the same time, the trend of wind turbine towards greater capacity and higher efficiency requires to improve the utilization rate of wind energy, which makes the study of maximum wind energy tracking of great significance. In this paper, the actual problems of wind power prediction and maximum wind power tracking are studied. For wind power prediction, because wind power and wind speed have a direct functional relationship, wind power prediction results can be obtained by predicting wind speed. Wavelet decomposition is used to reduce the nonlinear and uneven stability of wind speed series. Most of wind speed prediction is based on neural network theory, but there is no uniform method for selecting input quantity of neural network. In this paper, the method of selecting input quantity by time series modeling is used to select input quantity for neural network. To sum up, this paper presents a prediction method based on wavelet theory, time series and neural network, which is called wavelet time series neural network prediction method. In this method, the original wind speed is decomposed into a low frequency trend signal and several high frequency random signals by wavelet transform, and then the time series method is used to model the high frequency signal. Then, two kinds of signals after wavelet decomposition are modeled by BP neural network: the low-frequency signal is based on conventional BP neural network (the input is the most recent six historical values). The high frequency signal is constructed by time neural network (input quantity is modeled and selected by time series). Finally, the prediction results of each signal are reconstructed by superposition, and the final prediction results of the original wind speed series are obtained. The wind power can be predicted by inputting the wind speed prediction results into the fitted fan power curve. Based on the measured data of a wind farm in the south of China, the prediction method presented in this paper has good prediction accuracy. Aiming at the problem of maximum wind energy tracking, this paper studies the control method of variable speed constant frequency doubly-fed wind turbine based on the analysis of wind turbine operating characteristics. In order to simplify the difficulty of the model, the coordinate transformation of doubly-fed asynchronous generator is carried out, and the vector transformation technique of stator flux orientation is adopted. The simplified model can decouple the active and reactive power of the generator and achieve the goal of maximum wind power tracking. The simulation of wind power system with Matlab/Simulink is carried out, and the results show that the model is correct.
【学位授予单位】：华北电力大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TM614
本文编号：2388802
[Abstract]:With the decrease of energy sources, environmental pollution is becoming more and more serious, especially haze is gradually covering most cities in China. The use of clean and non-polluting energy has become an important choice to solve these problems. Because of its renewable, green, clean and environmental characteristics, wind power generation in the world is increasing. However, due to the intermittent and uncontrollable characteristics of wind power generation, the ability of wind power grid to accept wind power is restrained, and the development of wind power generation is slow. The prediction of wind power with high accuracy can effectively reduce the adverse effect of wind farm output fluctuation on power system and improve the competitive power of wind power generation in power market. At the same time, the trend of wind turbine towards greater capacity and higher efficiency requires to improve the utilization rate of wind energy, which makes the study of maximum wind energy tracking of great significance. In this paper, the actual problems of wind power prediction and maximum wind power tracking are studied. For wind power prediction, because wind power and wind speed have a direct functional relationship, wind power prediction results can be obtained by predicting wind speed. Wavelet decomposition is used to reduce the nonlinear and uneven stability of wind speed series. Most of wind speed prediction is based on neural network theory, but there is no uniform method for selecting input quantity of neural network. In this paper, the method of selecting input quantity by time series modeling is used to select input quantity for neural network. To sum up, this paper presents a prediction method based on wavelet theory, time series and neural network, which is called wavelet time series neural network prediction method. In this method, the original wind speed is decomposed into a low frequency trend signal and several high frequency random signals by wavelet transform, and then the time series method is used to model the high frequency signal. Then, two kinds of signals after wavelet decomposition are modeled by BP neural network: the low-frequency signal is based on conventional BP neural network (the input is the most recent six historical values). The high frequency signal is constructed by time neural network (input quantity is modeled and selected by time series). Finally, the prediction results of each signal are reconstructed by superposition, and the final prediction results of the original wind speed series are obtained. The wind power can be predicted by inputting the wind speed prediction results into the fitted fan power curve. Based on the measured data of a wind farm in the south of China, the prediction method presented in this paper has good prediction accuracy. Aiming at the problem of maximum wind energy tracking, this paper studies the control method of variable speed constant frequency doubly-fed wind turbine based on the analysis of wind turbine operating characteristics. In order to simplify the difficulty of the model, the coordinate transformation of doubly-fed asynchronous generator is carried out, and the vector transformation technique of stator flux orientation is adopted. The simplified model can decouple the active and reactive power of the generator and achieve the goal of maximum wind power tracking. The simulation of wind power system with Matlab/Simulink is carried out, and the results show that the model is correct.
【学位授予单位】：华北电力大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TM614
【参考文献】
相关期刊论文 前3条
1 徐大平;肖运启;吕跃刚;李炜;;基于模糊逻辑的双馈型风电机组最优功率控制[J];太阳能学报;2008年06期
2 田铭兴,励庆孚,王曙鸿;交流电机坐标变换理论的研究[J];西安交通大学学报;2002年06期
3 张新房,徐大平,吕跃刚,柳亦兵;大型变速风力发电机组的自适应模糊控制[J];系统仿真学报;2004年03期
相关博士学位论文 前1条
1 肖运启;双馈型风力发电机励磁控制与优化运行研究[D];华北电力大学（北京）;2008年
，本文编号：2388802
本文链接：https://www.wllwen.com/kejilunwen/dianlilw/2388802.html
教材专著
