Using smart metering and weather data for a small set of solar homes from NSW, this work develops preliminary machine learning models that provide forecasts of aggregate time-series net demand for regions of geospatially proximate houses. The work investigates strategies for building forecast models that do not rely upon near-real-time smart metering data nor fine-grained descriptive household data to produce reasonable estimates of future regional energy consumption.

The slide-deck highlights findings from the development and application of models based on the publicly accessible Ausgrid Solar Homes dataset. To enable other researchers to compare their own model performance against the preliminary CSIRO model, an interactive Tableau dashboard is also provided, highlighting model forecasting performance for Newcastle and Sydney regions for 2012-2013.