Online hierarchical forecasting for power consumption data

This paper proposes a three-step approach to forecasting time series of electricity consumption at different levels household aggregation. These are linked by hierarchical constraints—global is the sum regional consumption, for example. First, benchmark forecasts generated all using generalized additive models. Second, each series, aggregation algorithm ML-Poly, introduced Gaillard, Stoltz, and van Erven in 2014, finds an optimal linear combination benchmarks. Finally, projected onto coherent subspace ensure that final satisfy constraints. By minimizing regret criterion, we show projection steps improve root mean square error forecasts. Our tested on data; experimental results suggest successive