Generating peak-aware pseudo-measurements for low-voltage feeders using metadata of distribution system operators

Abstract

Distribution system operators (DSOs) face challenges such as restructuring distribution grids for climate neutrality and managing grid consumption and generation. Measurements within the grid are crucial for DSOs, yet many low-voltage (LV) grids lack measurement devices. To address this, an approach is proposed to estimate pseudo-measurements for non-measured LV feeders using regression models. The models are based on feeder metadata, which includes the number of grid connection points, installed power of equipment, and billing data in the downstream LV grid. The authors also incorporate weather, calendar, and timestamp data as model features and use the existing measurements as the model target. For evaluation, a dataset of 2323 LV feeders is used and peak metrics for magnitude, timing, and shape of consumption and feed-in are introduced, inspired by the BigDEAL challenge. The authors employ XGBoost, a multilayer perceptron (MLP), and a linear regression (LR) model, finding that XGBoost and MLP outperform LR. The results demonstrate that this approach effectively adapts to varying conditions and generates realistic load curves from feeder metadata. Additionally, the authors elaborate on feeders where pseudo-measurements exhibit deficiencies. This method could be extended to other grid levels such as substations and contribute to research in load modelling, state estimation, and LV load forecasting.