电力设备tanδ在线监测中的信号去噪

The dielectric loss factor tanδ is an important characteristic quantity of power equipment insulation condition. At present, the tanδ on-line monitoring devices of power equipments in china is widely exist the problems of low accuracy, stability, reliability, and some of them at risk of being shelved. The main reason of this phenomenon is that most of the existing de-noising algorithms or circuits can hardly to meet the requirements of the field with seriously noises, while the observables related to tanδ is so weak and the measurement requires a high accuracy. Based on the proposed de-noising theory of independent component analysis (ICA), this project will investigate the possibility of enhancing the performance of the ICA de-noising algorithms. Firstly, the theoretical defects of the ICA model will be researched in, and the influence of the extrinsic factors such as environment and samples, the ways of prior knowledge acquisition will be studied also. Then, an evaluation system will be established to analyze the feasible solutions to get a number of factors which influence the performance of the ICA de-noising algorithms. Finally, a new de-noising theory will be formed and a new algorithm can be got. The new algorithm can greatly improve the accuracy, reliability and stability of the tanδ on-line monitoring devices. Under the condition of SNR 6dB, the prospective parameter of the new algorithm is that the credibility of phased error below 0.06° is no less than 98%, while the credibility of phased error below 0.03° is no less than 90%.

电力设备介质损耗因数tanδ是表征电力设备绝缘状况的一个重要特征量。目前我国的电力设备tanδ在线监测装置广泛存在准确性、稳定性、可靠性不足，实用化水平不高的现象。其主要原因是由于电力设备现场噪声干扰严重，而tanδ的相关监测量很微弱且要求测量准确度较高，现有的去噪算法及电路大多难以满足实际要求。本课题以现有的独立分量分析（ICA）去噪理论为基础，从模型的理论缺陷、环境及样本等外在因素、先验知识获取三个方面探索提升ICA去噪算法性能的可能性，然后将所获得的可行方案输入一个评价体系，分析各自的优缺点并总结影响ICA去噪性能的若干关键因素，最终形成新的去噪算法和理论。新算法可大幅提升tanδ在线监测装置的准确度、可靠性和稳定性，预期信噪比6dB条件下，相位误差小于0.06°的可信度不低于98%；相位误差小于0.03°的可信度不低于95%。

介质损耗因素(tanδ)能够反映设备的早期缺陷，是表征容性电力设备绝缘性能的重要指标之一。本课题针对电力设备tanδ在线监测，重点开展强噪声背景下的微弱正弦信号去噪算法研究。课题首先从ICA去噪模型理论出发，分析出噪声成分的分离顺序与算法去噪准确度密切相关，提出了3种基于ICA的去噪算法：带约束的ICA二次法、基于SCA与FastICA的去噪算法和带约束的ICA排序算法。这3种算法分别从分离矩阵元素关系、分离矩阵初始值、源信号能量角度出发使得噪声成分首先被分离，在噪比6dB条件下，相位误差小于0.06°的可信度均达到了90%以上。课题接着提出了一种基于CCA与GSS的频率测量算法，通过准确测量被测信号的频率成分来自适应选择样本长度与起始点，设计出了一种适用于ICA去噪的数据预处理方法。最后，课题整合了上述提出的数据预处理算法以及3种基于ICA的去噪算法，以BP神经网络为基础，结合最大似然估计、负熵、拟合误差等评价量建立起了一个关于去噪准确度的评价体系，提出了一种基于ICA的多尺度正弦信号去噪算法。该算法在信噪比6dB条件下，相位误差小于0.06°的可信度不低于98%；在信噪比6dB条件下，相位误差小于0.03°的可信度不低于95%，单次平均计算时间不超过10秒。