M.S. AAI Capstone Chronicles 2024

ELECTRICITY DISTRIBUTION TOPOLOGY CLASSIFICATION

13

will group the timestamps by day for the pixels (24 pixels for each day), and then the LSTM design will be able to perform a sequence on the daily/weekly sequence. This approach will accomplish a similar temporal approach. However, the model will handle the sequencing through the LSTM design instead of the temporal aspect being handled as a pure image classification design.

Results & Conclusion

Baseline Model This project research began by establishing the baseline model, a crucial step in the project. Following Tennakoon, R.N. et al. 's methodology (Tennakoon, R.N. et al., 2020), Euclidean distance was calculated to measure similarity values between each smart meter's voltage profile and a reference meter from each transformer. Then, a support vector machine (SVC) was used for transformer prediction based on the smart meter similarity scores. After fine-tuning the SVM model at 0.65, the SVC model achieved final accuracy in weighted precision at 0.68 and recall at 0.65 on 262 transformer classifications. Comparatively, Tennakoon, R.N. et al.’s research, using the same method, achieved an accuracy of over 0.90 with 710 smart meters connected to five (5) transformers. This highlights the significance of the research, as it demonstrates that the data set used for this project, where 1258 meters were connected to 262 transformers, may not be able to reach high accuracy in the final result. However, the correlation between the meter voltage measurement and the transformer connection is still suitable for predicting the distribution

topology connections. Transformer Model

104