© Patrick Pollmeier / FH Bielefeld

Pow­er gen­er­a­tion fluc­tu­ates, so does con­sump­tion: solu­tion through arti­fi­cial intel­li­gence and edge computing

More and more pho­to­volta­ic sys­tems on rooftops, more and more e‑cars plugged into pow­er out­lets: this puts a strain on the pow­er grid due to fluc­tu­a­tions in gen­er­a­tion and con­sump­tion. The inter­na­tion­al research project “AI4DG” ini­ti­at­ed by Biele­feld Uni­ver­si­ty of Applied Sci­ences is now inves­ti­gat­ing how these fluc­tu­a­tions can be bal­anced out local­ly. The idea is to use dis­trib­uted AI to reli­ably and autonomous­ly con­trol the pow­er supply.

Biele­feld (fhb). On the roof of the Biele­feld Uni­ver­si­ty of Applied Sci­ences, they glis­ten and shine in the sun in such a way that you can lit­er­al­ly watch the mod­ules of the pho­to­volta­ic (PV) sys­tem gen­er­ate elec­tric­i­ty. “Unfor­tu­nate­ly, the sun does­n’t always shine so beau­ti­ful­ly,” says Katrin Schulte regret­ful­ly. With this, the spe­cial­ist for pow­er grids from the Uni­ver­si­ty of Applied Sci­ences (FH) Biele­feld names one of the two chal­lenges fac­ing grid oper­a­tors in the expan­sion of decen­tral­ized renew­able ener­gy gen­er­a­tion: Unlike con­ven­tion­al pow­er plants, a PV sys­tem does not gen­er­ate a con­stant and plannable out­put. “Weath­er-depen­dent or volatile pow­er gen­er­a­tion is what we call it,” Schulte explains. The pow­er gen­er­at­ed in this way is then fed into the low-volt­age grid to which pri­vate house­holds are also con­nect­ed. How­ev­er, these increas­ing­ly require a lot of pow­er, for exam­ple when charg­ing an e‑car. “That,” Schulte said, “is the oth­er big challenge!”

A decen­tral­ized AI to con­trol the pow­er grid.

Ener­gy gen­er­a­tion by many small plants and con­sump­tion in the low-volt­age range will there­fore increase in the future and at the same time fluc­tu­ate more strong­ly. “Con­trol­ling the net­work safe­ly becomes a com­plex and increas­ing­ly dif­fi­cult task due to the ini­tial sit­u­a­tion,” Katrin Schulte describes the prob­lem. That’s why the 26-year-old has helped launch an inter­na­tion­al research project to work out a promis­ing approach to the solu­tion: dis­trib­uted, or decen­tral­ized, arti­fi­cial intel­li­gence (AI) to con­trol the pow­er grid. Or as the project title says in Eng­lish: “Arti­fi­cial Intel­li­gence on the edge for a secure and autonomous dis­tri­b­u­tion grid con­trol with a high share of renew­able ener­gies (AI4DG)”.

Uni­ver­si­ties of Biele­feld and Greno­ble, Atos World­grid and West­falen Weser Netz also on board

The sci­en­tists involved start­ed work in Octo­ber. Katrin Schulte is involved as a doc­tor­al stu­dent for the UAS. The project builds on her pre­vi­ous research: After com­plet­ing her bach­e­lor’s degree in regen­er­a­tive ener­gies at Biele­feld UAS, the 26-year-old had stud­ied smart ener­gy sys­tems in depth in her mas­ter’s degree in elec­tri­cal engi­neer­ing. Since the begin­ning of 2020, she has been con­duct­ing research as a sci­en­tif­ic UAS employ­ee in projects on shap­ing the ener­gy tran­si­tion at the Insti­tute for Tech­ni­cal Ener­gy Sys­tems (ITES) in the Grids and Ener­gy Sys­tems (AGNES) work­ing group under the direc­tion of Prof. Dr.-Ing.

As ini­tia­tor and project leader, Prof. Haubrock super­vis­es AI4DG, in which part­ners with dif­fer­ent focal points col­lab­o­rate in a dig­i­tal­ly net­worked man­ner. While the idea, ini­tia­tive and pow­er grid exper­tise come from the Biele­feld Uni­ver­si­ty of Applied Sci­ences, the Uni­ver­sité Greno­ble Alpes con­tributes the AI know-how. Biele­feld Uni­ver­si­ty is tak­ing care of the already men­tioned decen­tral­ized dis­trib­uted use of AI, the so-called edge com­put­ing. Final­ly, indus­try part­ners Atos World­grid from France and the “Inno­va­tion — Intel­li­gent Grid Tech­nol­o­gy” divi­sion of West­falen Weser Netz from Ger­many are help­ing with imple­men­ta­tion in the field.

Edge com­put­ing: Gen­er­a­tion and con­sump­tion data are processed and ana­lyzed decentrally

AI is already being researched today in approach­es to con­trol­ling the pow­er sup­ply. But the Biele­feld research team and its French col­leagues are going even fur­ther and link­ing AI with edge com­put­ing. “Edge com­put­ing means that data is not processed cen­tral­ly in the cloud, but decen­tral­ly right where it is gen­er­at­ed,” explains Tim­on Jungh. The bio­mechatron­ics engi­neer (MA) is con­duct­ing research in the work­ing group of Prof. Dr.-Ing. Ulrich Rück­ert at CITEC (Cen­ter for Cog­ni­tive Inter­ac­tion Tech­nol­o­gy) and is also com­plet­ing his doc­tor­ate in the project. “This is a real inno­va­tion,” empha­sizes Katrin Schulte. “We are increas­ing secu­ri­ty and data pro­tec­tion with it, because pow­er sup­ply is part of the crit­i­cal infra­struc­ture and must be guar­an­teed.” If one decen­tral­ized AI unit fails, anoth­er can imme­di­ate­ly take over. And if data does­n’t have to be sent, it can’t be lost or hacked along the way. And there’s anoth­er advan­tage to the decen­tral­ized approach: “The data can be processed local­ly with less delay,” says Tim­on Jungh.

Bet­ter fore­casts can be made — sup­ply becomes more reliable

On site — that’s in Her­ford, for exam­ple: A large gray box with wide vents stands on the side of the road in a res­i­den­tial area. A yel­low stick­er with a black light­ning bolt is embla­zoned on the side: a local net­work sta­tion. It trans­forms incom­ing medi­um volt­age into low volt­age, which is then used to sup­ply the sur­round­ing hous­es. Mar­co Sawatz­ki opens the sta­tion. Cables, switch­es, coun­ters and a mea­sur­ing field are revealed. Sawatz­ki plugs in his lap­top. “Now you can see how much elec­tric­i­ty is cur­rent­ly being con­sumed by house­holds. Noth­ing seems to be fed into the grid at the moment,” explains the employ­ee of West­falen Weser Netz, the indus­try part­ner on the Ger­man side of the project.

AI4DG pro­vides that exact­ly at this point, direct­ly in the local sub­sta­tion, AI process­es the mea­sure­ment data. “We then used the AI analy­ses for fore­casts, for exam­ple. They are sup­posed to pre­dict the con­sump­tion of indi­vid­ual house­holds or the pow­er gen­er­at­ed by PVs,” explains Katrin Schulte. Depend­ing on this, the bat­ter­ies in the house­holds with PVs can then be con­trolled to serve the grid and the ener­gy gen­er­at­ed can either be stored or fed into the grid, so that there is always a con­stant volt­age and the pow­er sup­ply is secured and over­loads in the grid are avoided.

Pure trans­fer: from the lab, to the field, to the mar­ketable product

Before that can hap­pen, how­ev­er, Schulte and her col­leagues first devel­op and opti­mize their con­trol sys­tem on the com­put­er and in the lab­o­ra­to­ry: spe­cial soft­ware is used to set up an elec­tric­i­ty grid as a sim­u­la­tion, while West­falen Weser Netz sup­plies the real data for it. “This is our play­ground, this is where we try out our self-pro­grammed algo­rithms,” says Katrin Schulte. If they pass the tests, they go to the Smart Ener­gy Appli­ca­tions (SEAp) grid sim­u­la­tion lab. Real cur­rents flow here, and the con­trol sys­tem is val­i­dat­ed using var­i­ous hard­ware com­po­nents such as bat­tery stor­age and elec­tron­ic loads. “Only then will we test our sys­tem in a field tri­al in the real pow­er grid togeth­er with West­falen Weser Netz. And if it works well, the indus­try part­ners can devel­op our sys­tem into a mar­ketable product.”

Coop­er­a­tion with France enrich­ing — Europe-wide ben­e­fits sought

Pos­si­bly also for Europe-wide use: AI4DG is an inter­na­tion­al research project. It is fund­ed by the Ger­man Fed­er­al Min­istry of Edu­ca­tion and Research (BMBF) as part of the pro­gram for Fran­co-Ger­man projects on arti­fi­cial intel­li­gence with a total of around one mil­lion euros (the UAS will receive around 200,000 euros of this). Prof. Haubrock estab­lished the con­tact with the French part­ners and is pleased about the inter­na­tion­al coop­er­a­tion: “Through the coop­er­a­tion with France, dif­fer­ent dis­tri­b­u­tion grid struc­tures can be con­sid­ered for a trans­fer­abil­i­ty of the AI meth­ods to the Euro­pean ener­gy sys­tem. This will enrich the estab­lished, nation­al AI strategies.”