Prediction of Customer Connectivity to Distribution Transformer via Application of Machine Learning Techniques to Smart-Meter Data

Accurate topology information is essential for the efficient operation of a power distribution network. However, due to the constant changes in the physical topology, maintaining this accurate information is a challenging task. This research attempts to solve a common predicament faced by distribution network service providers (DNSPs), by providing an accurate connectivity model of the power distribution network. This would provide an accurate map as to which customer is powered by which distribution transformer. The dataset used for this study includes the average RMS voltage values from 710 smart meters powered by five transformers, every 15 minutes for five months. A classification model that can predict the smart meter connectivity to its respective transformer with an average accuracy of 90% is obtained. Initially, similarity values are calculated using the Euclidean distance measure by comparing the voltage profiles of each smart meter with that of a selected reference meter from each transformer. Next, this data is used to train the classifier using the algorithm created. Finally, an unknown set of smart meter data is tested using the trained model to observe the prediction accuracy. The solution provided could be further developed by adding features such as phase identification.