M.S. AAI Capstone Chronicles 2024

ELECTRICITY DISTRIBUTION TOPOLOGY CLASSIFICATION

7

smart meter's voltage profile and the reference voltage profile for each transformer, could resulting in a significant computational time. The dataset consists of individual hourly voltages (min, max, avg) for various meters corresponding to a designated transformer spanning over three months. The initial approach will include a CNN deep learning model to account for the change in power usage over the various periods. The CNN design effectively retains sequence information using a previous time step input, which fits our hourly data format through the location of pixels within the image. The approach of using a CNN Model for an image classification representation of the dataset was researched for plausibility and effectiveness. Sequence classification using CNN-LSTM networks can function by applying sliding convolutional filters to the input, which allows for the CNN to learn features from both the spatial and time dimensions (MathWay, 2024). An approach to classify and recognize the emotion of the speaker from the Database of Emotional Speech is to classify the individual spoken utterances, label them with their corresponding emotion, and plot them via spectrogram. The spectrograms were then used for the classification by the CNN to use the time and amplitude input. A similar approach was taken by an academic group at USD (McGirr et al., 2023), where the composer of an audio file was classified based on the sequential pattern of a particular instrument. In the notebook, the audio file was represented by selecting one instrument, mapping the intensity and timestamp of the note, and creating a spectrogram to be passed through a CNN/LSTM model to classify which spectrogram to the predicted composer. Following this similar approach, the model causes an intake of the power consumption, constructs a visual representation of the hourly power usage, and performs the image classification procedure for each meter grouped by the transformer.

98