【作者】 郭瑞；

【导师】 杨正瓴；

【作者基本信息】 天津大学 ， 模式识别与智能系统， 2010， 硕士

【摘要】 微网短期负荷预测对微网系统安全、经济的运行有重要的意义。微网短期负荷预测中,白噪声的存在导致了预测的准确率上限。如何从微网短期负荷序列中识别并分离白噪声,是尚未解决的难题。准确的分离白噪声,可以确定比较优化的预测模型与方法,确定组合预测模型的数目,得出概率化预测结果,对于微网短期负荷预测具有十分重要的意义。本文具体研究内容和结果如下:(1)采用Matlab进行了四种小波函数、软硬阈值函数下小波阈值去噪数值实验,验证了小波去噪与时间序列分析中的平滑法有相似之处,四种小波去噪效果的优劣与移动平均法和指数平滑法(接近于正态的)的好坏有某种对应关系,并从理论上推断构造更好的平滑系数的可能性。(2)采用解析形式探索了功率谱分离白噪声的可能性,并对AR(1)序列、AR(2)序列进行去噪研究。小波去噪和差分去噪按频率划分信号与噪声,对于低频信号高频噪声序列去噪有效,而功率谱去噪按幅值划分信号与噪声,对于序列是低频噪声高频信号的情形仍有效。在信号和噪声在频率域上不存在相互抵消的情况下,功率谱分离噪声方差效果显著。(3)在Smith和Wallis对预测之谜给出的经验解释的基础上,采用数理统计学进行了理论解释,研究了样本容量对组合权重的影响。(4)采用差分法和小波对eunite原始负荷序列的48个抽样序列进行去噪研究,发现差分法去噪对低频信号高频噪声序列的效果要优于小波去噪,验证抽样序列方差比原序列方差大。更多还原

【Abstract】 The research of microgrid short term load forecasting short-term is important for the safe and economic running of the system contains microgrid. For microgrid load, there exists an upper limit of forecast accuracy which is caused by unpredictable white noise. Generally speaking,it is an unresolved difficult problem to identify the distribution function of white noise and separate white noise from microgrid load time series .Accurate separation of white noise, can determine more optimal prediction model and more optimal prediction method, can determine the number of combination forecasting model, achieve probabilistic predictions, is of great significancefor the microgrid short term load forecastingThe main contents and results are:(1) Study the effect of wavelet denoise using four wavelet functions, hard and soft threshold function under Matlab simulation environment. Cofirm the similarities between wavelet denoising and the smoothing in time series analysis, the pros and cons of four types of wavelet denoising have something to do with the advantages and disadvantages between the moving average and exponential smoothing method which is close to normal, and infer the possibility of better smoothing coefficient in theoretically.(2) Explore the possibility of power spectrum denoising with analytical method, study the effect of power spectrum denoising using an AR(1) time series and an AR(2) time series.Wavelet denosing and differential method distinguish signal from noise by frequency, are usually used for the higher frequency noise on the lower-frequency signal sequence, while power spectrum distinguish signal from noise by amplitude, lower frequency noise on higher frequency signal sequence is still valid for power spectrum. For the case of signal and noise do not offset in frequency domain, the effect of power spectrum denoising is remarkable.(3) On the basis of empirical discovery of Smith and Wallis, give a mathematical statistics analytical explanantion for the forecast combination puzzle. Study the sample size’s effect on the combined weights.(4) Study the effects of differential method and the wavelet for 48 sample load sequence from eunitel load sequence. Find differential method denoising is superior to wavelet denoising for higher frequency noise on lower frequency signal sequence. Confirm sample sequeces have larger variance.更多还原