超短期风电预测及考虑风速预测的惯性控制研究

发布时间：2018-02-27 04:42

本文关键词： 风电出力特性风速功率曲线超短期风电预测时间序列法智能算法组合预测自适应神经模糊推理系统惯性控制　出处：《西南交通大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着风电在电力系统中的渗透率越来越高,其固有的不确定性给电网安全经济运行带来严峻挑战。对风电出力概率特性进行统计分析,实现高精度的超短期风电预测能为电力系统运行管理人员提供应对风电不确定性的基础条件。风电预测精度越高,电网接纳风电的能力越强。另一方面,越来越多的风电机组代替传统的同步电机,使系统频率响应特性持续恶化,因此风电机组参与系统调频在工业界和学术界均得到广泛重视。本文基于四川地区某实际风电场的实测数据,对风电出力特性、超短期风电预测和惯性控制进行研究,详细内容如下:1、对该风电场年出力分布和年风速分布进行统计分析,对比15min、30min和60min时间尺度下的出力波动特性;统计分析该风电场风电机组之间出力的相关性和互补性;分析单台风机在同一风速下输出功率的宽范围分布现象,提出了一种基于最优平滑阶数的风速功率曲线建模策略,以最优平滑阶数处理原始风速得到输入风速从而建立风速功率曲线模型,并与已有方法中以原始风速为输入建立的风速功率曲线模型进行精度对比分析;2、提出了一种基于BP神经网络的长时间尺度缺失风电功率数据的补齐方法,以缺失数据时间段内另7台风机的实测风电功率数据为BP神经网络的输入,得到待补齐风电机组的功率数据,并与常用的相邻风机法进行精度对比分析;3、深入研究基于历史功率数据的超短期风电预测,实现了持续法、ARMA、ARIMA这三种时间序列法;在考虑风电功率序列波动特性的基础上提出了一种改进的持续法,其精度较持续法有一定程度的提升;实现了 BP神经网络、RBF神经网络、SVM、PSO-SVM四种智能算法。对上述预测方法在不同季度以及不同预测步长情况下的预测结果进行对比分析。选择时间序列方法中总体精度最好的ARMA方法和智能算法中总体精度最好的PSO-SVM为子预测方法,利用ANFIS组合上述两个子预测方法从而得到最终的风电预测结果,并将组合法的预测精度与两个子预测方法进行对比分析;4、分析了暂态过程中风速波动对风机参与系统调频的影响,进而提出了一种基于超短期风速预测的惯性控制策略。基于未来10s平均风速的预测值来设计ROCOF和droop控制环的增益且每10s更新一次增益。在仿真系统中,基于多种来自实测数据的风速波动情况,开展了包括切机和负荷跃升等扰动的算例研究,对比分析了基于超短期风速预测的惯性控制策略与恒定增益的惯性控制策略的控制效果。
[Abstract]:With the increasing permeability of wind power in power system, its inherent uncertainty brings severe challenges to the safe and economic operation of power grid. The realization of ultra-short-term wind power prediction with high accuracy can provide basic conditions for power system operation managers to deal with the uncertainty of wind power. The higher the precision of wind power prediction, the stronger the power grid's ability to accept wind power. More and more wind turbines are replacing the traditional synchronous motors, which make the frequency response characteristics of the system deteriorate continuously. Therefore, the participation of wind turbine in FM system has been paid more and more attention in industry and academia. Based on the measured data of a practical wind farm in Sichuan area, the characteristics of wind power generation, prediction of wind power and inertial control are studied in this paper. The detailed contents are as follows: 1. The annual output force distribution and the annual wind speed distribution of the wind farm are statistically analyzed, and the fluctuation characteristics of the output force are compared under the time scale of 15 min or 30 min and 60 min respectively, and the correlation and complementarity of the output force among the wind farm units are analyzed statistically. Based on the analysis of the wide range distribution of the output power of a single typhoon under the same wind speed, a modeling strategy of wind speed power curve based on the optimal smoothing order is proposed. The input wind speed is obtained by processing the original wind speed with the optimal smoothing order, and the wind speed power curve model is established. The accuracy of the wind speed power curve model based on the original wind speed input is compared with that of the existing methods, and a new method based on BP neural network is proposed to correct the wind power data. Taking the measured wind power data of the other 7 typhoon turbines in the missing data period as the input of BP neural network, the power data of the wind turbine to be compensated are obtained. By comparing and analyzing the accuracy of the conventional adjacent fan method, the ultra-short-term wind power prediction based on historical power data is studied in depth, and the three time series methods, ARMA-ARIMA, are realized. On the basis of considering the fluctuation characteristics of wind power series, an improved persistence method is proposed, the accuracy of which is improved to a certain extent. Four intelligent algorithms of BP neural network and RBF neural network are implemented. The prediction results of the above prediction methods in different seasons and different prediction steps are compared and analyzed. The overall accuracy of the time series method is selected. The best ARMA method and the best overall precision of the intelligent algorithm are PSO-SVM subprediction methods. The final wind power prediction results are obtained by combining the above two sub-prediction methods with ANFIS. The prediction accuracy of the combined method and the two sub-prediction methods are compared and analyzed. The influence of the wind speed fluctuation on the frequency modulation of the fan participating in the system is analyzed in the transient process. Furthermore, an inertial control strategy based on ultra-short-term wind speed prediction is proposed. The gain of ROCOF and droop control loops is designed based on the predicted value of the average wind speed of 10 s in the future and the gain is updated every 10 s. Based on the fluctuation of wind speed from measured data, a numerical example of disturbance, such as cutting machine and load jump, is carried out. The control effects of inertial control strategy based on ultrashort wind speed prediction and constant gain inertial control strategy are compared and analyzed.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM614

【参考文献】