【作者】 高阳；

【导师】 杨正瓴；

【作者基本信息】 天津大学 ， 模式识别与智能系统， 2009， 硕士

【摘要】 科学的预测是正确决策的依据和保证。微网短期负荷预测是电力系统领域的一个前沿的重要研究课题,对含微网系统运行的安全性和经济性有重要的意义。对于一个特定的电力短期负荷记录(时间序列),存在预测准确率的上限。对微网负荷,预测准确率上限的大部分由白噪声的不可预测性引起。因此,研究负荷序列中的白噪声分离,具有重要的实际意义。一般地,从复杂时间序列中识别白噪声的正确分布函数,是尚未解决的困难理论问题。本文结合微网负荷预测,从理论上分析了白噪声与信号的相对性;采用小波等方法,研究了白噪声近似分离。依此进行了微网短期负荷模拟数据的白噪声分离研究。具体研究内容和结果如下:(1)结合微网负荷的特点,提出了微网负荷预测的“背景调整法”,并对背景形成的思路做了初步的探讨。(2)从理论上分析了有限长度(样本容量)时间序列中信号与噪声的相对性。根据Cramer分解定理,可将非平稳时间序列分解为信号和白噪声两部分。发现对于有限长度时间序列,趋势项和随机项的区分是相对的,白噪声和信号的区分是相对的。并通过小波去噪等数值仿真证实:样本容量越大,这种区分越准确。大体符合卡方分布规律。(3)得到了一个理想的白噪声序列,其性能显著优于用“伪随机数”生成的白噪声序列。(4)采用白噪声、AR(1)序列,评估了小波去噪的效果。并采用Bior4.4、Sym8、Db4、Coif5小波,对微网负荷模拟序列进行了白噪声分离。发现在功率谱上信号与噪声可分条件下,小波去噪有一定的实际价值。更多还原

【Abstract】 Scientific forecasting is the foundation and assurance of correct decision. Short-term load forecasting of microgrid system is a leading subject, which is important to the economy and stability of microgrid systems.For a given load data series(a time series), there exists an upper limit of forecast accuracy. For microgrid load, most of the upper limit is caused by white noise, so the study of the denoise is of practical signifcance. Generally, to identify the distribution function of white noise from a complex time series is an unresolved difficult problem.Here we study the relativity of noise and signal, combined with the microgrid load forecast; Study the white noise separation using strategies as the wavelet denoise, and apply the wavelet denoise to microgrid load.The main contents and results are:(1)According to the characteristic of microgrid, put forward a“background adjust strategy”, and study how to construct an ideal background.(2) Study theoretically the relativity between signal and noise in limited time series.According to Cramer’s decomposition theorem, a given non-stationary time series can be decomposed into two uncorrelated different parts: signals and white noise. For a time-limited series, the difference between trend and random is relative, the difference between noise and signal is relative. Find that if the sample capacity is larger, the difference is bigger.(3)Achieve an ideal white noise by processing a physical white noise. Its performance is much better than that achieved from pseudo-random number.(4)Study the effect of wavelet denoise using white noise and an AR(1) time series. Apply wavelet denoise to microgrid load using wavelets like bior4.4, sym8, db4 and coif5. Find that when the power spectrums of signal and noise are separable, the wavelet denoise is very effective.更多还原