【作者】 李鹏；

【导师】 杨洪耕；

【作者基本信息】 四川大学 ， 电力系统及其自动化， 2004， 硕士

【摘要】 目前对电能质量的监测已经从传统的按照国家标准考核电能质量统计指标发展到一个新的阶段，电力部门和电力用户逐步要求对动态电能质量信号进行记录和分析。传统电能质量监测仪不能记录电能质量暂态波形，国外高精度的监测装置又无法进行长期的电能质量监测，因此本文提出了一种采用数据压缩和消噪技术的解决方案，采用该算法不但可以在不增加硬件投入的基础上满足对动态电能质量事件的长期监测，同时可以显著降低通讯成本，达到对整个电能质量监测网的实时监测。该算法不但能够向电能质量管理中心提供直观的电压、电流波形，而且使管理者能够使用重构信号进行识别、定位、量化分析、分类以及查找事件原因。 动态电能质量扰动信号通常包含较宽的频谱，因此电能质量监测仪需要采用较高的采样频率，这造成数据存储和传输成本增加，必须进行数据压缩；同时由于采样数据常叠加了系统白噪声以及测量白噪声，因此也必须对信号进行噪声抑制。电能质量扰动信号类型繁多，并且信号噪声水平不同，所以本文提出了具有数据自适应性的信号压缩和消噪算法。 由于对某一个信号最好地变换对另外一个信号来说不一定是最佳的，因此本文采用小波基和局部余弦基构造了一个完备的变换模型库，从而可以从模型库中选择针对特定信号的最佳变换，实验证明各类以及各种噪声水平的信号都能在模型库找到稳定的信号描述。本文通过引入Beylkin算法和divide一conquer算法等技术很好的解决了离散小波变换和局部余弦变换的时间平移不变性问题。算法对信号进行局部余弦变换时采用动态规划算法计算最佳局部余弦树，通过获得信号在时间轴上的最佳划分来获得信号的最优嫡编码。 本文采用信息论的MDL判据进行信号模型的估计。针对不同噪声水平和不同信号类型，这种算法具有数据自适应能力，不需要进行任何先验的参数设置 (例如阀值设置)和主观判断就能确定保留分解系数的最佳个数，并能根据信号动态选择模型，计算实例表明该算法能够满足电力部门的要求，同时图表证明了MDL能够在信号保真度与信号压缩效率之间找到最佳的契合点。 本文设计了一种信号模型选择的辅助判据“压缩比和保真度平衡策略”，该判据结合MDL，通过平衡“压缩比”和“均方误差”两个矛盾目标函数，能够选择最优的变换模型以及针对重构信号的最佳分解系数的个数。 通过和目前常用的4种小波数据压缩和消噪算法的对比分析可以看出本文算法综合性能指标优异，其很好地保留了电能质量扰动信号的时频特征，取得了令人满意的效果。关键词:动态电能质量扰动离散小波变换局部余弦变换数据压缩消噪时间平移不变性动态规划MDL更多还原

【Abstract】 The monitor of power quality has gone through the traditional stage of assessing statistic indicator of power quality comparing national standards to a new stage, electric utility industry and customers want to record and analyze dynamic power quality disturbance. The traditional power quality monitors can’t record transient wave of power quality, foreign monitor equipments with high precision can’t monitor power quality continually, so this paper proposes a new scheme for data compression and de-noising technique, using this algorithm not only continually monitor dynamic power quality events without increasing any additional hardware investment, but also decreasing communication cost remarkably, so the objective of real-time monitor for power quality network is achieved. This algorithm can supply power quality management center with visual voltage and current waves, at the same time, the managers can use these reconstructed signals to detection, localization, quantification, classification and events cause searching.Generally, the frequency range of DPQD signals is very band, so the sampling frequency may be very high for the accurate analysis, which will bring a large data storage and communication cost, so it’s necessary to process data compression; simultaneously, the sampling data is usually overlapped by additive system and measure white Gaussian noise, so noise suppressing is also obligatory. There are many sorts of DPQD signals, and their noise levels are different, so this paper proposes a data-adaptive algorithm for data compression and de-noising technique.The best transform to compress a particular signal may not be good for anothersignal, therefore, instead of restricting our attention to a particular basis, we consider a library of bases. A self-contained model library is made up of wavelet bases and local cosine bases. This collection of bases is highly adaptable and versatile for representing various DPQD signals.This algorithm has solved the time translation-invariant problem of DWT and LCT by introducing such techniques as Beylkin algorithm and divide-conquer algorithm and so on. In this algorithm, dynamic programming is used to calculate optimal local cosine tree during the LCT for signals, and the optimal entropy code is gained by the best partition of time axes.The information theory-based minimum description length criterion is used to estimate signal model. This algorithm has data-adapted capability for different noise levels and different sorts of signals. It can not only select the "best" number of retained decomposing coefficients without any parameter setting (such as thresholding) or subjective judgments, but also choose the "best" transform model according to signals dynamically. Computer results show that the algorithm can satisfy electric utility industry’s desire, simultaneously, illustrations prove that MDL criterion gives the best compromise between the fidelity of signal and the efficiency of signal compression.This paper design an assistant criterion (compression ration and error of signal reconstruction balancing strategy) for signal model selection, this criterion with MDL aims to gain the compromise between the two opposite object function of CR and MSE (mean square error). This criterion selects the best t signal model and the best number of coefficients for the signal reconstruction.Compared with four prevalent data compression and de-noising algorithms, analysis results show that integrated performance indicators of this algorithm is excellent. Time and frequency characteristics of DPQD are reserved perfectly and its result is satisfying.更多还原