Seamless RES Forecasting
Generic Seamless Forecasting Approach for Multiple Time Scales
Partner: ARMINES
Actors involved: RES producers, Aggregators
Context |
The prediction of RES production over multiple time scales currently consists in applying separate forecasting tools for different horizon intervals due to the fact that different phenomena explain predicted production over the horizon range.
An incremental approach is proposed to generate forecasts either at the plant level or at the level of an aggregation of plants, with coherent forecasts throughout the hierarchy of the aggregation.
Summary |
The production of a single RES plant or of an aggregation of RES plants is forecasted by a generic model, adaptive to several energy sources (e.g. Wind, Photovoltaics, Run-of-River Hydro) and able to predict multiple time scales at once (e.g. from 1 min to 2 days) without the need to re-train or re-configure the model at each horizon.
This generic seamless forecasting model offers to end-users of RES production forecasts such as RES producers or aggregators the possibility to simplify their forecasting process by optimizing their decisions over a large horizon range and several energy sources without discontinuities.
Challenge |
The prediction of RES production over multiple time scales results in a complex modelling chain that creates discontinuities at the junctions between intervals.
So how can the generic model have a continuous and equal performance compared to the advanced benchmark models? And can it simultaneously adapt to different energy sources over the range of horizon intervals?
Approach |
The generic seamless forecasting approach was developed by using Generic Analog Ensemble (PV, Wind, PV+Wind) which has a derivation of optimal weights, compared to other approaches.
A single generic seamless model substitutes at least 6 advanced models specific to particular horizon intervals and energy sources with same performance.
Innovative content of forecasting solution
The original contributions of this solution are:
– Continuous forecasting performance over the total horizon range.
– Streamlined forecasting process with a reduction of at least 50% in the number of required forecasting models.
– A generic model, adaptive to numerous variable renewable energy sources.
KPI Evaluation of RMSE and CRPS in the context of generic seamless RES forecasting.