Load Pattern Recognition Method for Probabilistic Short-Term Load Forecasting at Low Voltage Level

Advanced distribution management systems, aiming to determine the uncertainties in future demand and thus facilitate the management of low voltage networks, require an accurate approximation of vast amounts of streaming time series data generated by smart meters. This paper proposes a new load pattern recognition method for reducing the size of the load data model while preserving the information needed to support the probabilistic short-term load forecasting at the low voltage level. The proposed method recognizes the clusters of streaming time series and approximates the cluster members with representative central moments at forecast time resolution. The central moments extracted at the cluster level are used to quantify the future load uncertainties at the level of individual consumers. The usefulness of the proposed method is proven in a case study on the real US and Australian smart meter data. The case study results confirm that the new method is convenient for subsequent probabilistic forecasting.