基于小波图像卷积的局部放电非梯度定位方法研究

Non-gradient Localization of Partial Discharging Based on Wavelet Image Convolution

为实现更准确的电力设备及线路异常放电定位,提出了一种基于小波图像卷积的非梯度定位方法。研究首先采用小波图像卷积的时延估计方法解决了低信噪比下时延估计误差较大的问题,基于小波变换分析获得了信号间时域–频域相关性,进一步通过卷积平均值获取精确时延估计值,在真实值±2μs的估计结果区间内将时延估计的置信概率提升了44%。在较高信噪比下应用时,有效地将估计结果的平均绝对误差限制在0.1μs内;其次,将下山单纯形法和模式搜索法2种非梯度优化方法引入局部放电定位算法中,借助非梯度优化算法无需求导、搜索迅速等优点,实现了局部放电源的定位寻优。为验证上述算法效果,首先通过仿真对比验证了小波图像卷积时延估计方法在不同时延长度、不同信噪比下的准确性和稳定性,然后采用十字阵列超声传感矩阵,对模拟放电源进行了实测。实验结果表明,所提出的定位算法相较传统算法误差更小,且未增加对硬件算力的要求。

To achieve a more accurate location of the abnormal discharging in the power equipment and lines,this paper proposes a non-gradient localization based on the wavelet image convolution.Firstly,the research adopts the wavelet image convolution to solve the large delay estimation errors under a low signal-to-noise ratio.Based on the wavelet transform analysis,the time domain-frequency domain correlation between the signals is obtained.The accurate delay estimation value is further obtained through the convolution average value.The confidence probability is improved by 44%within the true value±2μs.When the proposed time delay estimation method is applied at higher signal-to-noise ratios,the mean absolute error is limited to 0.1μs.Secondly,the two non-gradient optimization methods of the Nelder-Mead method and the Hooke-Jeeves method are introduced into the partial discharge localization algorithm.The partial discharge is located by the non-gradient optimization algorithms with their advantages of being without solving the derivation but searching in rapid speed.To verify the effect of the above algorithms,this paper compares the wavelet image convolution with the traditional methods firstly.The accuracy and stability of the wavelet image convolution are verified.Then the cross-array ultrasonic sensing matrix is used to measure the simulated discharging source.The experimental results show that the proposed positioning algorithms have lower errors than the traditional methods without increasing the requirement of computing power.