Skip to main content

Forecasting

The analysis of a Low Voltage (LV) network often treats its demand and its violations as known: they are clustered, ranked, recommended and screened after the fact. A distribution network operator must act before the outcome is known. It switches, curtails, charges and reinforces on the strength of what it expects the load, the generation and the voltage to be. This part forecasts them.

Chapter 1 treats the signals in the notation of the book. It treats what is forecast in an LV network and for which decision, how a series is characterised before any model is fitted, how much of it can be forecast at all, and how inputs are chosen without leaking the future. Chapter 2 treats the forecasters: which families of models forecast the series, and how two forecasters are compared so that a conclusion survives the noise of training. Chapter 3 does the same for probabilistic forecasts and their evaluation. Chapters 4 and 5 apply the foundations where the forecast is hardest, the household, and where the operator acts, the transformer. Chapter 6 makes the uncertainty of a forecast trustworthy, and Chapter 7 values forecasts by the decisions that they inform.

Every signal in the part is measured. The AusNet households supply demand, a measured profile of photovoltaic output and recorded charging of electric vehicles. The customers of Porsgrunn in Norway supply hourly records of consumption and production together with the network behind each transformer. Each chapter is accompanied by a notebook that computes every figure it contains.