Electricity-demand forecasting
Probabilistic forecasting of local demand and the state's annual peaks, informing battery and solar use.
Role: Casual Academic Researcher, Adelaide University, January to June 2025. I worked in a four-person team with two software engineers and a tech lead, for a start-up client.
The problem
The client needed demand forecasts to decide when users should charge batteries, and when to draw on stored and solar energy, so they rely less on the grid during peak hours.
What I built
- I designed and implemented the forecasting algorithm, combining NeuralProphet, Gaussian processes and probabilistic fusion of multiple forecasting approaches.
- It forecasts local electricity demand up to two weeks ahead.
- It also forecasts the state’s top 10 annual demand peaks.
- I built the evaluation and benchmarking.
How the work was split
I set milestones with the client and the tech lead, who set the final goals and managed the project and schedule. The start-up’s engineers collected the data and implemented the interface to the algorithm.