Feeder load composition tracking for smart metered low voltage circuits

As Distributed Generation penetrates Low Voltage networks in greater quantities, the behaviour of the loads on these circuits, typically small and residential customers, must be better understood to avoid unnecessary investment and capitalize on increased efficiencies. Estimating thermal constraints as well as islanding capabilities hinges on accurate and representative load profiling, which requires periods of typical behaviour to be gathered through metering. The increasing availability of Smart Meters offers a solution to this with high frequency load measurements that align with generation dispatch periods. Prior work has taken steps to develop finer grained load profiles than those developed at the national level but given the high propensity for variability in residential customers, such metrics are overly general for small power systems. This paper takes previous work on load profiling at MV and LV levels and uses it to generate load profile compositions for learning the composition of customers that make up an LV feeder load and how it evolves over time. A simplification of a residential load profile model is applied to a set of real AMI data on a simulated feeder, resulting in 3 categories of user, a stratification learned from historical metering data. This yields an abstracted disaggregation of the loads on the feeder that accommodates the high variability and the heterogeneous profiles within the residential loads. Since compositional data is defined over the simplex rather than a real space, this restricts the statistical tools available for modelling to an inflexible subset. The solution presented circumvents this problem by utilising transforms that map the contributions to the aggregated feeder load into real space permitting a wider selection of analysis tools to be applied. A demonstration using a ye