Learning Estimators in order to Optimize the Management of Energy
Résumé
We deal here with the joint management of a photovoltaic platform and of the consumption of resulting energy by jobs subject to temporal constraints. Running a job requires assigning it a battery, loaded by the platform with some amount of energy, and this interaction through the batteries between the energy production activity of the platform and the consumption of this energy by the jobs requires the implementation of complex synchronization mechanisms. Since it most often involves distinct players with their own agenda and non-shared information, we short-cut the production level with the help of surrogate estimators. Those estimators involve pricing mechanisms and machine learning devices. We design and test several algorithms that implement this approach.