EDISON

Within EDISON, we develop and demon­strate AI-driven solu­tions for the effi­cient manage­ment of distrib­uted energy storage systems. We design dedi­cated proto­cols and data manage­ment systems to enable battery storage to work within Virtual Power Plant (VPP) frame­works. We iden­tify the following key project objec­tives:

• Develop an adap­tive AI-powered VPP manage­ment solu­tion: we build a smart, inter­op­er­able plat­form that coor­di­nates distrib­uted battery storage systems in real time. Edge-based AI algo­rithms handle moni­toring and opti­miza­tion locally, while secure commu­ni­ca­tion proto­cols and stan­dard­ized inter­faces keep the system connected. At the local level, deep Rein­force­ment Learning (RL) agents learn optimal charging and discharging poli­cies directly from inter­ac­tion with the energy envi­ron­ment, avoiding the need for hand-crafted control rules. For coor­di­na­tion across multiple sites, we employ Multi-Agent RL (MARL) so that each site's agent can learn coop­er­a­tive strate­gies while respecting local constraints. To preserve data privacy, we inte­grate Feder­ated Learning (FL) and Feder­ated Distil­la­tion (FD), enabling collec­tive model improve­ment across sites without trans­fer­ring sensi­tive oper­a­tional data. The plat­form adapts dynam­i­cally to changing grid condi­tions and envi­ron­mental factors, with shielding mech­a­nisms ensuring RL agents always remain within safe oper­ating bound­aries.
• Estab­lish AI-based battery life­cycle manage­ment: we develop intel­li­gent methods to esti­mate battery State of Health (SoH) and Remaining Useful Life (RUL) using a self-adap­tive approach that combines physics-based battery models with data-driven AI. These hybrid SoX esti­ma­tors contin­u­ously refine their predic­tions as oper­a­tional data accu­mu­lates, extracting reli­able health indi­ca­tors from field data without requiring disrup­tive full-discharge cycles. FL is used to improve esti­ma­tion models across distrib­uted sites, allowing knowl­edge transfer between hetero­ge­neous battery systems while preserving commer­cial confi­den­tiality.
• Enable smart market inte­gra­tion: we develop tools that allow VPPs to partic­i­pate effi­ciently in energy markets, including day-ahead, intraday, and reserve markets.
• System Inte­gra­tion and Vali­da­tion: we build and test the proposed solu­tions through both virtual simu­la­tions and real labo­ra­tory demon­stra­tors. A dedi­cated VPP simu­la­tion envi­ron­ment serves as a testbed for eval­u­ating RL algo­rithms and FL coor­di­na­tion proto­cols under real­istic grid condi­tions before phys­ical deploy­ment.

This re­search project is funded by the “Energy re­search: exploiting poten­tial and shaping the future” program of the Climate- and Energy funds owned by the Republic of Austria, oper­ated by the Federal Ministry for Climate Protec­tion, Envi­ron­ment, Energy, Mobility, Inno­va­tion, and Tech­nology (Project No. 926774).