Model-Free Voltage Calculations for LV Networks with LCTs Using Deep Neural Networks

As the penetration of low-carbon technologies (LCTs) continues to increase in the low voltage (LV) distribution network, distribution network operators (DNOs) need to ensure that the voltages remain within the accepted statutory limits. This is traditionally done by using electrical model-based simulation-driven approaches that are time-consuming and dependent on some error-prone assumptions. Model-free voltage calculation methods provide a faster simulation while maintaining accuracy. This paper evaluates the ability of electrical model-free voltage calculations to scale as the penetration and diversity of LCTs such as electric vehicles EVs and photovoltaic (PV) systems increase on the distribution network. Three scalable model-free algorithms are developed using deep neural networks (DNNs) and validated using real smart meter data (SM) for a realistic 31-customer feeder. The overall results show that the model-free method can accurately calculate voltages with average deviations of less than 1 volts and 2 volts for 25% and 50% LCT penetration scenarios, respectively.