GRIDouble - Next Generation Energy operational digital twins
GRIDouble is a comprehensive energy management tool that completely automates the finding of optimal patterns in energy consumption and production in the case of facilities with renewable energy sources.
The introduction of renewable energy sources (RES) in the grid has posed several challenges to energy producers and consumers. All these challenges change the role of Distribution System Operators (DSO) from energy transmitters into “orchestrators” of a large number of prosumers, which differ in size, type, and patterns of consumption and production. Orchestration and optimization of such complex energy systems using traditional methods is no longer feasible. These challenges require the application of novel approaches based on the synergy of IoT and AI.
Based on the data acquired during the grid exploitation, GRIDouble automatically creates the most adequate predictive models of the energy production and consumption, which are additionally improved by using publicly available data on weather, working days, electricity prices, solar irradiation, etc. To secure that the generated predictive models are always genuine digital twins of the installed equipment and the prosumer’s habits, the models are recreated periodically, thus evolving together with the energy ecosystem. Once we have the models that can accurately predict energy production and consumption, we can simulate any realistic or hypothetic operation plan and asses its effects. The results obtained from the simulations are subjected to an optimization process to find the energy management pattern that will result with the most economical usage of energy under the given conditions.
The innovativeness of our solution is twofold and is reflected in solving two pressing problems in the application of machine learning: the small amount of data, and the complexity of the implementation of the ML solution.
The biggest challenge in machine learning today is insufficient amount and quality of available data. For that purpose, we introduced Physics Informed Neural Networks (PINN) as a viable solution for training deep neural networks with few training examples, for cases where the available data is known to respect a given physical law, as in the case of i.e., PV production. Knowledge of an underlying physical law can enable neural networks to generalize well even when only a few training examples are available.
To solve the complexity of implementing ML solutions, GRIDouble utilizes Blackfox, the Vodena’s platform that automates the ML pipeline through the seamless integration of AutoML and MLOps components.
This project (GRIDouble) has received funding from the I-NERGY Project. The I-NERGY Project has received funding from the European Union’s Horizon 2020 Research and Innovation programme under grant agreement No. 101016508.