风电随机波动的方差预报研究

发布时间：2018-02-21 05:34

本文关键词： 风速方差物理特性预测误差分析小波分析　出处：《哈尔滨工业大学》2015年硕士论文　论文类型：学位论文

【摘要】：对风速进行预报是确保大规模风电安全并网的重要技术手段。但是目前预报的时间分辨率为15min的风速是一种平均意义下的风速,而真实风速由平均风速和风速的瞬时随机波动两部分构成,风速的瞬时随机波动的研究还未被关注。在平均风速预报的基础上对风速瞬时随机部分进行研究,可以提供实时风速的更加详细的信息,有助于电网系统制定更为详尽的平抑风电功率波动的策略。此外在风机的选型及安全设计时,风速的瞬时随机波动也是一个重要的因素。因此本文将研究的视角投向风速瞬时随机波动,主要的研究工作如下:首先本文定义了风速瞬时随机波动的方差这一概念来研究风速的瞬时随机波动,并基于小波分解理论提出了风速方差的计算方法。同时对风速方差的物理特性进行了研究。研究发现:风速方差与平均风速之间存在多尺度调制效应;风速方差的日变化曲线存在明显的日周期现象。电力系统在平抑风电功率波动时需要提前掌握未来一段时间内的风速信息,因此对风速方差进行预报具有重要的意义。对风速方差进行预报的前提是其具有可预报性。因此本文基于时间序列分析中的相关分析理论,提出了风速方差可预报性分析的方法。分析结果表明风速方差的可预报长度大致在1h至5h之间,可以进行预报。同时对风速方差与平均风速的依赖关系进行了研究。目前对于预报模型的最佳输入并没有很好的理论指导,因此本文通过实验选取预报模型的风速方差最佳输入维数。利用内蒙古风电场2013年风速方差数据分别进行提前10min、提前20min、提前30min、提前40min、提前50min以及提前60min时刻的风速方差预报。选择MAE及MSE作为预报结果的评价指标。同时利用黑龙江风电场2013年的风速方差数据的验证表明所建立的模型具有良好的稳定性和可推广性。在原始预报模型的基础上,本文通过加入平均风速的信息,改善了模型的预报性能,提高了预报结果的精度。最后,本文对风速方差模型的预报误差进行了统计分析。随着预报步长的增大,预报误差也随之增大,因此在预报的过程中,预报步长的选择至关重要,否则会使预报结果的可信度降低。利用带位移和尺度的T分布、正态分布以及极值分布对预报误差的分布进行拟合,结果表明预报误差的分布最符合带位移和尺度的T分布。而且不论是用BP神经网络进行预报还是使用支持向量机进行预报,预报误差的分布都与带位移和尺度的T分布最接近。
[Abstract]:The prediction of wind speed is an important technical means to ensure the safety of large-scale wind power grid, but the wind speed with 15min temporal resolution is an average wind speed. But the real wind speed is composed of two parts: average wind speed and instantaneous random fluctuation of wind speed. The research on instantaneous stochastic fluctuation of wind speed has not been paid attention to. The instantaneous stochastic part of wind speed is studied on the basis of average wind speed prediction. It can provide more detailed information on wind speed in real time, which is helpful for power grid system to develop more detailed strategy to stabilize the fluctuation of wind power. In addition, in the selection and safety design of fan, The instantaneous stochastic fluctuation of wind speed is also an important factor. The main research work is as follows: firstly, this paper defines the concept of the variance of the instantaneous stochastic fluctuation of wind speed to study the instantaneous random wave of wind speed. Based on the wavelet decomposition theory, the calculation method of wind speed variance is proposed, and the physical characteristics of wind speed variance are studied. It is found that there is a multi-scale modulation effect between wind speed variance and average wind speed. The diurnal variation curve of wind speed variance has obvious daily periodic phenomenon. In order to stabilize the fluctuation of wind power, the power system needs to master the wind speed information for a period of time in the future. Therefore, it is of great significance to predict the variance of wind speed. The premise of forecasting the variance of wind speed is its predictability. Therefore, based on the theory of correlation analysis in time series analysis, A method for analyzing the predictability of wind speed variance is proposed. The results show that the predictable length of wind speed variance is between 1 h and 5 h. At the same time, the dependence between wind speed variance and mean wind speed is studied. At present, there is no good theoretical guidance for the best input of the prediction model. In this paper, the best input dimension of wind speed variance of prediction model is selected by experiment. Wind speed in advance of 10 minutes, 20 minutes, 30 minutes, 40 minutes, 50 minutes and 60 minutes in advance of Inner Mongolia wind farm on 2013, respectively. Velocity variance prediction. MAE and MSE are selected as the evaluation indexes of forecast results. Meanwhile, the verification of wind speed variance data of Heilongjiang wind farm on 2013 shows that the established model has good stability and extensibility. Based on the prediction model, In this paper, the prediction performance of the model is improved by adding the information of the mean wind speed, and the precision of the forecast result is improved. Finally, the prediction error of the wind speed variance model is analyzed statistically. The prediction error also increases, so the choice of the prediction step is very important in the forecast process, otherwise, the credibility of the forecast result will be reduced. Using the T distribution with displacement and scale, Normal distribution and extreme value distribution fit the distribution of prediction error. The results show that the distribution of prediction error is most consistent with T distribution with displacement and scale, and it can be predicted either by BP neural network or by support vector machine. The distribution of prediction error is the closest to the T distribution with displacement and scale.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TM614

【参考文献】