Abstract
Partial discharge diagnosis from offline field tests conducted on power cables are notably onerous due to a multitude of reasons - signal attenuation, frequency dis persion, different cable lengths, multiple joints and noise influences. Such complex ity causes categorisation conflicts when deterministic rules are applied; the same an tecedent may map to different consequents. Deep learning, in contrast, offers a data based design that is probabilistic in nature. The formulation of machine-driven inter pretation and feature selection has been proven to be successful when used in tandem.The works presented here administers a multi-step artificial intelligence approach iterated with examination of extracted features for partial discharge identification, lo calisation and generation. To evaluate the performance of the algorithms, case stud ies are deliberately selected to comprehensively delineate the difficulties encountered during field testing. The experimental results justify the capabilities of the proposed concepts in identifying, localising and generating discharges besmirched with signif icant quantities of noise. Main contribution is the successful automated diagnosis of measurements acquired under challenging field constraints