Adaptive templates for matched filter bank for continuous online partial discharge monitoring

Matched filtering can be applied to recorded signals for maximum detection sensitivity during online partial discharge (PD) measurements in medium-voltage cable circuits. PD waveform predictions are combined with a noise spectrum estimation to construct matched filters. The PD waveform is specific for each cable circuit under test and the predictions are based on online system identification measurements and a standard cable model. For some cable circuits, for instance with a ring-main-unit along the cable, the predicted waveforms only approximate the real waveforms due to signal distortion not accounted for in the model. This results in suboptimal detection sensitivity. This paper describes an automated procedure for updating the matched filters based on detected PD signals. The algorithm analyzes detected pulses to judge whether the current set of matched filters perform optimally. If not, the current set of matched filters is updated, increasing the PD detection sensitivity and potentially also the location accuracy. At the same time, recognition of recurring disturbances is improved and better rejection criteria can be defined. The method is tested on signals measured on live cable connections.