Decentralized Model-Free Voltage Calculation Using Neighboring Smart-Meter Data

The increasing penetration of distributed energy resources (DERs) challenges voltage regulation in low-voltage (LV) distribution networks, where detailed network models and feeder-wide measurement infrastructure are often unavailable. Model-free voltage calculation based on machine learning (ML) offers an alternative by exploiting smart-meter data. However, existing ML-based approaches are centralized and require smart-meter data of all customers, limiting their applicability when full information is unavailable. This paper proposes a decentralized model-free approach that uses a deep neural network (DNN) to calculate the voltage of a target customer using only its own smart-meter data and a limited number of neighboring customers. The proposed approach requires neither knowledge of the network topology nor access to feeder-wide measurements, reflecting practical real-world conditions. Its performance is evaluated for different numbers of neighboring customers under varying PV and EV penetration levels. Results show that larger neighborhoods generally improve voltage calculation accuracy. This highlights a trade-off between voltage calculation accuracy and the level of customer data sharing, while indicating that customer selection can influence accuracy. However, accuracy decreases at higher PV and EV penetration levels.