M.S. AAI Capstone Chronicles 2024

ELECTRICITY DISTRIBUTION TOPOLOGY CLASSIFICATION

4

interval format for readings has been deemed more suitable, and a KNN approach has been

selected to fill in missing data points. Additionally, instances of multiple readings within the

same hour have been noted. A new aggregation method has been developed to standardize these

readings into consistent hourly intervals.

Secondly, the dataset includes three types of voltage measurements: average, minimum,

and maximum voltages recorded within a 60-minute window. To streamline the data handling,

these types are now consolidated into a single timestamp entry with three new corresponding

columns. Some entries deviated from the expected measurement types during the initial data

exploration. After consultation, these entries were identified as errors and have been excluded

from the analysis to maintain dataset integrity.

Upon completing the processing and analysis of the datasets, 1258 meter files suitable for training were obtained, each tagged with a corresponding ground truth transformer connection label. Additionally, there are 294 meter files lacking these labels, which SaskPower will utilize as a "final testing" set upon submission of the results. The analysis of the 1258 files revealed several key findings. Firstly, the distribution of meter voltage readings follows a normal distribution, as illustrated in Figure 2, which suggests a standardized variability across the data. Secondly, each transformer voltage reading displayed a unique trend in the magnitude of the meter readings, as shown in Figure 3. This characteristic is crucial as it can help classify meter topology more accurately. Adding time-sequence data to the voltage magnitude, as suggested by the trends shown in Figure 3, could enhance feature sets for topology classification, aiding in better model performance. Last but not least, the dataset is relatively balanced, as demonstrated in Figure 4; most transformer labels are associated with 4-5 meters and, in some instances, 8-9 meters. This

95