The Medium Voltage (MV) network in the Netherlands is almost entirely composed of underground cables. The failure statistics show that the interruptions in the medium voltage have a high contribution in the mean outage time per year per customer. The interruptions in the medium voltage grid are often caused by failure of circuit components mainly the cables, joints, and terminations which account for 73% of the failures. The occurrence of these interruptions and their duration can be limited by proper maintenance measures. Partial discharge diagnostics provide a way for the condition assessment of the circuit insulation. The Smart Cable Guard (SCG) systems from DNV GL help in continuous monitoring of the medium voltage cable circuits during the circuit operation and aid the network operators in maintaining the MV grid. With its non-intrusive monitoring characteristic, we see that the SCG systems provide the measurement of partial discharges (PD) occurring in the cable circuit. Moreover, the partial discharge measurements consists of the information about the location of each observed discharge event along with the time of its occurrence and its corresponding discharge magnitude observed in picocoulombs for every minute. The partial discharge events occurring in the circuit are subjected to evaluation from an expert at DNV GL control rooms who qualitatively assigns a warning level (Level 1, 2, 3, or Noise) to the observed pattern of discharge events. This manual process of evaluation of PD events and assignment of warnings to them is a laborious task and the network operator has to rely on the availability of the expert and his/her accurate assessment. To reduce this dependence for the network operator, Alliander in collaboration with the DNV GL experts is developing an automated decision support tool to identify the partial discharge events and to aid the operator and the expert in evaluating the condition assessment of the cable network. This thesis proposes a clustering methodology using the ST-DBSCAN density-based technique for identifying high-density discharge events or `areas of interest' in the PD data obtained from the SCG systems. The clusters identified from the method are further evaluated by extracting their features or characteristics using the PD data attributes as well as describing their
characteristics based on the context of the circuit. The performance of the clustering method is validated using the DNV GL warnings by formulating performance indicators and metrics to measure the performance. The clustering method along with the features extracted from the cluster contribute towards the development of the automated decision support tool.