In most of the cases, currently installed smart meters are being used only to collect and store data on aggregated end-user consumption for billing purposes. Although smart meters in the distribution network also collect processing data that include phase voltage measurements, additional possibilities are not completely utilized, which makes analyzes such as consumption forecasting and distribution network state estimation even more difficult to carry out. Due to the complex features of the collected time series data, the traditional methods and applied models are no longer suitable for network analyses, and the implementation of more complex algorithms, often based on machine learning, is required. This paper presents algorithms that determine the accurate phase connectivity of end-users and calculate their phase consumption distributed by phases from the collected phase voltage measurements and aggregated end-user consumption. After the phase consumption is determined, consumption and losses forecasting in the observed low-voltage network is performed. Such analyses, in addition to being suitable for determining additional features from the available data, serve as a basis for detecting anomalies and technical and non-technical losses in distribution networks.

Croatian Science Foundation call DOK-2021-02