Supervisor:
Josep M. Guerrero

Co-Supervisor:
Juan C. Vasquez

Najmeh Bazmohammadi

Assessment Committee:
Claus Leth Bak (Chair)
Juan Manuel Roldan Fernández, Universidad de Sevilla
Ramon Blasco-Gimenez, Technical University of Valencia

Moderator:
Sanjay Chaudhary

Abstract:

Researchers have been focusing on comprehending brain theories and taking cues from natural evolution and human intelligence. The goal is to develop devices, techniques, and methodologies to tackle societal problems. The synergy between neuroscience and AI has enhanced our understanding of language, cognitive processing, learning strategies, sensory systems, and neural processes. Scientific, industrial, and societal improvements have all benefited from neuroscience insights.
While computational neuroscientists have developed several neural circuit models, there is still limited validation of these models outside of neuroscience, mainly in energy systems. Developing machine learning (ML) algorithms that can continuously learn in real-life scenarios and manage dynamic datasets for industrial applications remains an arduous task. Though AI-based solutions have demonstrated promise in several energy systems, current approaches face issues with biological realism, energy efficiency, high computing complexity, flexibility, and limited online learning capabilities.
Microgrids (MGs), characterized by their decentralized architecture and complex control systems, are a paradigm shift in the industrial and energy sectors. To provide autonomous and self-sufficient energy systems and facilitate optimal operation management, these intricate energy systems integrate a variety of components, such as cyber-physical control devices, smart energy storage units, and distributed energy resources (DERs). By smoothly switching between isolated and grid-connected modes, MGs can dynamically modify their operating mode, improving power supply efficiency and resilience. Their intrinsic flexibility, however, creates specific challenges that call for various monitoring and control techniques to guarantee trustworthy operation, especially in the face of variable voltages, currents, and frequencies. Furthermore, uncertainties remain regarding their bidirectional power flow, fault current limitations, and dynamic behavior.
The purpose of this PhD thesis is to investigate how neuroscience-inspired algorithms could potentially overcome these obstacles and support MGs to reach their full potential. First, it examines the fault location techniques in Smart Grids (SGs) to identify challenges, requirements, and potential solutions. Second, it analyzes current communication systems and AI-based methods for protecting MGs, focusing on networked microgrid protection strategies. Third, since the integration of an AC MG in a distribution system causes coordination issues in the protection system, an adaptive protection strategy is suggested to address these issues. Inspired by the remarkable automation and self-defense mechanisms in human brains, this thesis also explores different strategies to overcome the shortcomings of AI-based solutions.
This study explores the application of fundamental and theoretical amygdala brain functions as well as the predictive coding inference process to develop innovative AI strategies for MG protection that are founded in neuroscience. It evaluates the application and challenges of brain-inspired learning strategies, such as emotional learning, in the field of MGs. The Conclusion section summarizes the thesis outcome and makes recommendations for potential future research directions.

 

Resumé:
Forskere har i de seneste år fokuseret på at forstå hjerneteorier og har taget inspiration fra naturlig evolution og menneskelig intelligens. Målet er at skabe ressourcer, strategier og metoder til at løse samfundsproblemer. Vores viden om sprog, kognitiv behandling, indlaeringsstrategier, sensoriske systemer og neurale processer er blevet forbedret takket være integrationen af neurovidenskab og kunstig intelligens. Neurovidenskabelige indsigter har hjulpet videnskabelige, kommercielle og samfundsmæssige fremskridt.
Computational neuroscientists har lavet mange neurale kredsløbsmodeller, men der er kun lidt validering uden for neurovidenskab, især i energisystemer. Det er stadig en udfordrende opgave at udvikle algoritmer for maskinlæring (ML), der kan lære kontinuerligt i virkelige scenarier og håndtere dynamiske datasæt til industrielle applikationer. Biologisk realisme, energieffektivitet, høj computerkompleksitet, fleksibilitet og begrænsede online læringsmuligheder er problemer, som de nuværende tilgange står over for, selvom AI-baserede løsninger har vist sig at være lovende i flere energisystemer.
Microgrids (MG's) er et paradigmeskifte i både industri- og energisektoren på grund af deres decentraliserede arkitektur og komplekse kontrolsystemer. Disse sofistikerede energisystemer består af en række komponenter, herunder cyberfysiske kontrolenheder, smarte energilagringsenheder og distribuerede energiressourcer (DER's), for at yde optimal og autonom energistyring. MG's er i stand til dynamisk at ændre deres driftstilstand ved regelmæssigt at skifte mellem nettilsluttede og isolerede tilstande. Dette forbedrer strømforsyningens effektivitet og modstandsdygtighed. Men deres naturlige fleksibilitet giver nogle problemer, der kræver en række overvågnings- og kontrolmetoder for at sikre pålidelig drift selv under variable spændinger, strømme og frekvenser. Derudover er deres tovejsstrøm, fejlstrømsbegraensninger og dynamiske adfærd stadig usikre.
Denne ph.d.-afhandling undersøger, hvordan neurovidenskabsinspirerede algoritmer kan overvinde disse problemer og hjælpe MG's med at realisere deres fulde potentiale. Først ser den på fejllokaliseringsmetoderne i Smart Grids (SG's) for at finde problemer, mulige løsninger og fremtidige arbejder. For det andet fokuserer vi på kommunikationsstrategier og AI-baserede metoder til at beskytte MG's. For det tredje, efter at en AC MG er blevet integreret i et distributionssystem, foreslås en adaptiv beskyttelsestilgang til at løse de koordinationsproblemer, der opstår i beskyttelsessystemet. Denne forskning undersøger også forskellige måder at overvinde AI-baserede løsninger på grund af de bemærkelsesværdige automatiserings- og selvforsvarsmekanismer i menneskelige hjerner. Denne undersøgelse undersøger, hvordan man bruger grundlæggende og teoretiske amygdala-hjernefunktioner samt prædiktive coding-inferensprocessen for at udvikle innovatie AI-strategier til MG-beskyttelse, der er baseret på neurovidenskab. Den evaluerer anvendelsen og udfordringerne ved hjerneinspirerede læringsstrategier, såsom følelsesmæssig læring, inden for MG'er. Afslutningsafsnittet opsummerer afhandlingens resultater og giver anbefalinger til mulige fremtidige forskningsretninger.