Clustering

Every distribution network serves customers whose demand differs, and almost every decision a Distribution System Operator (DSO) takes about them, from a tariff to a connection approval, treats them in groups. This part asks how such groups can be found in smart-meter data, how they can be validated when no labels exist to check them against, and what they contribute to the operation of the Low Voltage (LV) network.

The chapters build on one another. Chapter 1 introduces the definitions of a cluster, the representations of a load curve and the instruments for judging a partition, and illustrates each on a synthetic population whose structure is known in advance. Chapter 2 applies these instruments to three metered populations, settles a clustering pipeline, and establishes the practice of reading every index together with the cluster sizes, the archetype curves and an external attribute. Chapter 3 applies the pipeline to the AusNet population and assesses whether the resulting archetypes are reliable enough to support decisions that persist. Chapter 4 places the same households on a real network and asks whether archetypes carry information about network risk, and Chapter 5 evaluates the uses of clustering in network studies that retain the level of demand. Chapters 6 and 7, in preparation, examine representations that retain the magnitude of demand and the clustering of customers on a physical feeder.

Three metered populations recur throughout the part, summarised below and described in Table 2.1 of Chapter 2. The Low Carbon London trial supplies 1,284 households and is the main population of Chapter 2, where a cohort of households whose overnight peak is the signature of storage heating provides the one structure that can be verified independently. The Spanish energy cooperative GoiEner supplies 1,000 households, and the AusNet distribution network in Victoria supplies 342 residential households metered every half-hour for a year. The AusNet households also form the demand pool of four published Australian Medium Voltage (MV)-LV network models, which allows Chapters 3 to 5 to place the households that are clustered on the urban network HPK11 and to simulate its response to rooftop Photovoltaic (PV) and Electric Vehicle (EV) charging. Each chapter is accompanied by a notebook that computes every figure and table it contains.

Population Country Households Reading interval Used in
Low Carbon London United Kingdom 1,284 30 min Chapter 2
GoiEner Spain 1,000 60 min Chapter 2
AusNet Australia 342 30 min Chapters 2 to 5