复杂电力扰动信号的盲源分离方法研究

Conventional power quality studies have laid the theoretical foundation for the power disturbance signal analysis, however, the capabilities of existing approaches to complex disturbance signal analysis are still to be improved. A general purpose method to extract useful information from complex and diverse power disturbance signals is both theoretically and practically important to promote the use of power quality disturbance data for the other applications, such as fault diagnosis. The project proposes blind and semi-blind source separation methods for complex power disturbance signal analysis, which can be applied to the applications with/without prior knowledge of the objectives. The research firstly studies the typical characteristics of complex power disturbance signals and investigates on the selection of objective function of the blind source separation model. How to incorporate prior knowledge, from different perspectives, into the model as constraints are discussed. A new optimization algorithm combined the heuristic algorithm and the traditional Newton termed as two-particle algorithm is developed to ensure optimized convergence and convergence speed. In order to solve the single channel blind source separation problem, the morphological filter are designed to build an virtual multi-channel observation data matrix; and finally design a prototype of power disturbance monitoring and analysis device based on system-on-chip and validate the effectiveness of the proposed algorithm by experiments. This study will provide new methods and theoretical basis for the analysis of complex power disturbance signals.

传统的电能质量分析研究工作为电力扰动信号的分析奠定了理论基础，但现有方法对复杂扰动信号的分析能力仍有待提高。研究更通用的分析方法，从复杂多样的扰动信号中提取有用信息，对推动扰动数据在电能质量之外的其他领域中的应用，如故障诊断，具有重要的理论和现实意义。本项目在盲源分离理论的统一框架下，提出复杂电力扰动信号的盲/半盲分离方法，分别适用于没有/有先验知识两种情况。首先，对复杂电力扰动信号的特征进行分析，研究盲源分离模型的目标函数的选取机制，从不同角度探讨将先验知识转化为约束条件的方法，并设计结合自启发式算法与传统牛顿法的双粒子优化算法，保证优化的收敛性和收敛速度；针对单通道观测信号的欠定盲源分离问题，深入研究形态学滤波器的设计，构建有效的虚拟多通道观测数据矩阵；最后，基于片上系统技术进行电力扰动监测和分析装置原型的设计和算法的实验验证。本研究将为复杂电力扰动信号分析提供新的方法和理论基础。

电力扰动中蕴含了大量涉及电力系统和设备运行状态的有用信息，通过信号处理技术对这些信息进行提取和辨识，对电力系统和设备的安全稳定运行具有重要的意义。然而，在某些场景下的电力扰动信号呈现出复杂的混叠特征，使得直接对信号进行特征提取和参数识别变得困难。本项目提出了基于盲源分离理论的电力扰动分析方法。项目首先对不同场景下的电力扰动信号的混叠特性和统计特性进行了分析，明确其可分离性。在此基础上选择适用的盲源分离方法，同时分析盲源分离方法的不确定性及其对所解决问题的影响。提出了基于数学形态学的信号预处理方法，对电力扰动信号进行起止时刻定位、滤波消噪，使得观测信号能更好地满足可分离条件。针对只有一个观测信号的极端情况，采用延时嵌入或多尺度形态滤波的方法，构造虚拟观测矩阵，从而把问题转化为MIMO的多通道盲源分离问题，降低了欠定盲源分离问题的求解难度。将盲源分离方法应用于低频振荡、电能质量复合扰动等信号的分离和参数辨识，通过和其他信号分解方法进行对比测试，验证了项目所提出的方法具有更好的分离效果。目前，已发表期刊/国际会议论文2篇(其中SCI收录2篇，EI收录3篇)，已申请专利4项（授权发明专利1项）。