„Without Data, AI is worthless.“

---

Artificial intelligence is already helping doctors with diagnoses, cars can drive autonomously thanks to its support, and AI is even writing software autonomously. Will AI soon be controlling our power grids fully automatically?

Prob­a­bly not, since we are still in the begin­ning. And in crit­i­cal infra­struc­ture in par­tic­u­lar, it is still usu­al­ly peo­ple who make the final deci­sion. This is also the case in med­i­cine. But Arti­fi­cial intel­li­gence will grad­u­al­ly play an ever greater role in mas­ter­ing the increas­ing com­plex­i­ty of our net­works. And it is already being used in many areas. At MR, we have been work­ing with machine learn­ing for 20 years. It all start­ed with our dehy­drat­ing breathers. That was the first time we imple­ment­ed self-learn­ing algo­rithms for con­trol. We have grad­u­al­ly com­plet­ed our sen­sor port­fo­lio and equipped our prod­ucts with intel­li­gent algo­rithms. In par­tic­u­lar our solu­tions for eval­u­at­ing oper­at­ing resources are becom­ing increas­ing­ly com­pre­hen­sive and pre­cise thanks to AI. How­ev­er, the poten­tial is far from exhaust­ed.

What is the greatest potential of AI solutions?

In view of the tech­no­log­i­cal advances in com­put­er and data pro­cess­ing, cur­rent­ly par­tic­u­lar­ly in large lan­guage mod­els, there seem to be no lim­its. In the near future, I see the great­est advan­tages in the sta­tus assess­ment of the var­i­ous assets in the net­work and in net­work man­age­ment. How­ev­er, for AI to be able to exploit its advan­tages, it first needs a lot of data – with­out data, AI is worth­less. Com­pre­hen­sive data col­lec­tion for trans­form­ers, cir­cuit break­ers and cables is there­fore essen­tial. To make this pos­si­ble, we are work­ing on reduc­ing the costs of dig­i­tal­iza­tion.

How would you like to reduce the costs of digitalization transformers?

Arti­fi­cial intel­li­gence helps us here too. At the moment, I can’t tell you too many details, but we are cur­rent­ly test­ing algo­rithms that draw con­clu­sions from the mea­sure­ment data of a few sen­sors and cal­cu­late val­ues for oth­er fic­ti­tious or dig­i­tal sen­sors. The soft­ware com­pen­sates for the func­tion of oth­er sen­sors, so to speak, by intel­li­gent­ly com­bin­ing the mea­sure­ment data with his­tor­i­cal data and eval­u­at­ed mod­els. This enables us to assess the con­di­tion of trans­form­ers cost-effec­tive­ly and effi­cient­ly.

Regarding more efficient condition assessment: Can this already be realized today?

ETOS^® is already estab­lished on the mar­ket at field lev­el for mon­i­tor­ing and col­lect­ing data from trans­form­ers. With its Asset Intel­li­gence func­tion and our DGA inter­pre­ta­tion, for exam­ple, we already offer data and mod­el-based solu­tions for more effi­cient con­di­tion assess­ment includ­ing rec­om­mend­ed actions for oper­a­tors. That trans­form­ers them­selves will soon be active­ly con­tact­ing asset man­agers for con­di­tion-based main­te­nance or nec­es­sary repairs is obvi­ous. Mes­sages will then come either from the oper­at­ing sys­tem on the trans­former – such as our ETOS^® – or from a dig­i­tal twin in an asset per­for­mance man­age­ment sys­tem such as TESSA^® APM.

Can AI also help to optimize grid management?

Def­i­nite­ly, that’s the future! We are already work­ing on such a solu­tion togeth­er with a part­ner. Real-time con­di­tion assess­ment of var­i­ous trans­former assets, cir­cuit break­ers and cables in com­bi­na­tion with his­tor­i­cal data and, for exam­ple, weath­er fore­casts, offers immense poten­tial. One exam­ple here is proac­tive trans­former cool­ing in order to with­stand upcom­ing pow­er peaks or over­loads with a min­i­mized reduc­tion in ser­vice life; and this is just the begin­ning. To some extent, this is already a real­i­ty, but in the future AI will play a much greater role in net­work man­age­ment.

How can you ensure that AI is used safely in critical infrastructure?

Any­one who uses Chat­G­PT knows that answers can also be wrong because AI some­times hal­lu­ci­nates a lit­tle. Although the results seem plau­si­ble, they are fac­tu­al­ly wrong. Of course, this must not hap­pen with our AI solu­tions and mod­els, which have to be one hun­dred per­cent reli­able. To train our algo­rithms, we there­fore only use data that we have col­lect­ed and processed our­selves in field tests or that has been made avail­able to us by net­work oper­a­tors for these pur­pos­es.

Every step, from label­ing the data to check­ing its plau­si­bil­i­ty and test­ing, is car­ried out by our own experts. We have been work­ing with trans­form­ers and their com­po­nents for sev­er­al decades and know exact­ly what we are doing. This exper­tise is embed­ded in our algo­rithms. Because our prod­ucts are used in crit­i­cal infra­struc­ture, they are also sub­ject to the new EU AI Act which impos­es par­tic­u­lar­ly strict rules on the han­dling of data and the use of arti­fi­cial intel­li­gence in gen­er­al. Of course, we com­plete­ly com­ply with all known require­ments of the EU AI Act.

What happens to customer data?

Quite sim­ply: it stays with the cus­tomer. Nei­ther MR nor third par­ties have access to cus­tomer data, either at the field lev­el with ETOS^® or in the Asset Per­for­mance Man­age­ment Sys­tem with TESSA^®. To this end, we have imple­ment­ed an ISMS infor­ma­tion secu­ri­ty man­age­ment sys­tem based on ISO 27001 to ensure that data pro­tec­tion and IT secu­ri­ty are guar­an­teed for our sys­tem solu­tions dur­ing devel­op­ment and, of course, lat­er dur­ing oper­a­tion. But of course we are always hap­py to receive data from real net­work oper­a­tions because this makes our AI solu­tions even bet­ter which also pro­vides ben­e­fits to our cus­tomers.

4 examples of how artificial intelligence can help energy suppliers

---

AI made by Reinhausen

---

The fol­low­ing prod­ucts and solu­tions with AI tech­nol­o­gy are already avail­able.