Increasing penetration of low carbon technologies in residential low voltage (LV) networks increases the need for modelling their state to preempt voltage issues. Due to the challenges in modelling vast numbers of feeders, LV network models are often simplified, incomplete, or even absent. The largescale roll-out of smart meters (SMs), creates the opportunity for generating accurate LV network models at scale at low cost. In this paper, a methodology for voltage estimation in LV networks without an electrical model is proposed and tested across 127 real distribution feeders. The approach uses machine learning and historical active power and voltage data from SMs to predict voltage at a node of interest. This approach shows promising results, particularly in its capability to generalise to different loading scenarios and estimate voltage under higher electric vehicle and solar photovoltaic penetration levels.

Demand response programmes are essential for the stability of a power system with increasing penetration of renewable generation. However, before the necessary technologies are deployed, the flexibility potential that consumers can offer to the grid operator must be evaluated. Although smart meters can support this process, aggregated consumption profiles do not provide insights into the individual appliances. Non-intrusive load monitoring techniques are an alternative, but most recent works only focus on increasing the performance of primarily supervised learning algorithms, reducing the applicability of their results. This paper proposes an end-to-end methodology to quantify flexibility potential and emissions intensity of large household appliances. The process relies solely on local aggregated measurements and low-complexity unsupervised learning techniques to identify major power blocks. This information is combined with information on wholesale market prices and electricity production mix to derive three novel metrics, dubbed peak price, peak emissions, and critical peak score. The results show that relatively simple unsupervised techniques can help characterise major consumption blocks from aggregated consumption, providing the per appliance flexibility potential.