Echo state network based smart meter data-driven short-term voltage forecasting for future power grids

Future power grids are expected to host high penetrations of low-carbon technologies such as rooftop photovoltaics (PVs), electric vehicles (EVs), and energy storage systems. The increasing integration of these technologies can introduce significant voltage deviations in low-voltage (LV) distribution networks. This paper proposes a data-driven, real-time voltage forecasting method based on Echo State Networks (ESNs), chosen for their computationally efficient recurrent neural network architecture, enabling fast training. Unlike many deep learning approaches that require metaheuristic optimization or signal preprocessing to improve forecasting, the proposed ESN model uses only smart meter data, specifically, active and reactive power measurements, together with prior voltage estimates, to forecast node voltages one step ahead. The model is evaluated using one-minute resolution smart meter data from a real residential household equipped with both PV and EV systems within an LV feeder. Results show that the ESN achieves high forecasting accuracy, with most errors within ±2 V, and significantly improves computational efficiency, averaging just 0.03 seconds per training run. These advantages make the ESN a promising tool for enabling dynamic grid management strategies in future distribution networks.