Loading...
Publication
Harnessing digital twin technology for smart grid optimisation in renewable energy communities
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 7.42 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
A transição global para sistemas energéticos sustentáveis exige quadros metodológicos
avançados capazes de gerir a complexa dinâmica estocástica inerente às Comunidades de
Energia Renovável (CER). Os atuais paradigmas de simulação mostram-se insuficientes para
capturar as interações multi-escalar entre recursos energéticos distribuídos, comportamento
dos prossumidores e restrições da rede. Esta dissertação apresenta avanços significativos para
a plataforma de simulação CityLearn, uma plataforma de código aberto para aprendizagem por
reforço multiagente na gestão energética de edifícios, através de cinco inovações
metodológicas fundamentais que, em conjunto, elevam a sua fidelidade em direção aos
padrões de Digital Twin: (1) adaptação da granularidade temporal permitindo comutação
dinâmica de resolução em intervalos infra horários, (2) integração de novos modelos físicos
incluindo eletrodomésticos como máquinas de lavar, (3) otimização multiobjetivo que
reconcilia metas de sustentabilidade e económicas concorrentes, (4) realismo estocástico
através de modelação probabilística de incertezas na geração e procura, e (5) serialização
padronizada de dados para interoperabilidade.
Uma revisão sistemática baseada na metodologia PRISMA identifica primeiro as limitações
críticas das abordagens de modelação de CER existentes, particularmente no que diz respeito à
sua representação de incertezas do mundo real e fenómenos emergentes à escala comunitária.
O desenvolvimento subsequente segue uma metodologia de engenharia formal
compreendendo: (1) síntese de requisitos a partir de estudos de caso operacionais de CER, (2)
melhoria arquitetónica da estrutura de simulação, e (3) validação abrangente contra conjuntos
de dados sintéticos e empíricos.
A validação experimental num caso de estudo com 17 edifícios demonstra as capacidades da
estrutura através das novas funcionalidades desenvolvidas: redução de 73,5% nas emissões de
carbono através de estratégias de controlo otimizadas para energia solar, redução de custos
em 71% em edifícios com objetivos de otimização diferenciados, e manutenção dos parâmetros
de conforto térmico dentro de limiares de ±0,5°C, tudo isto enquanto se alcança uma melhoria
de 18,6% nas métricas de estabilidade da rede. Estes resultados fornecem evidência
quantitativa para a otimização simultânea de objetivos de sustentabilidade, económicos e de
fiabilidade em operações de CER.
As contribuições metodológicas proporcionam valor imediato para operadores de comunidades
energéticas, enquanto estabelecem bases essenciais para o desenvolvimento futuro de Digital
Twins. Para os decisores políticos e económicos, os resultados empíricos oferecem benchmarks
verificáveis para o desenho de CER, particularmente no equilíbrio entre metas ambientais e
restrições operacionais. Ao avançar as capacidades de simulação em direção aos padrões de
Digital Twin, este trabalho fornece um elo crucial em falta entre os modelos teóricos de
transição energética e os requisitos de implementação no mundo real.
The global transition toward sustainable energy systems necessitates advanced methodological frameworks capable of managing the complex, stochastic dynamics inherent in Renewable Energy Communities (RECs). Current simulation paradigms remain insufficient for capturing the multi-scale interactions between distributed energy resources, prosumer behaviour, and grid constraints. This dissertation presents significant advances to the CityLearn simulation platform, an open-source platform for Multi-Agent Reinforcement Learning (MARL) in building energy management, through five key methodological innovations that collectively advance its fidelity toward Digital Twin standards: (1) temporal granularity adaptation enabling dynamic resolution switching across sub-hourly intervals, (2) expanded asset integration incorporating household appliances including washing machines, (3) multi-objective optimisation reconciling competing sustainability and economic goals, (4) stochastic realism through probabilistic modelling of generation and demand uncertainties, and (5) standardised data export serialization for interoperability. A systematic PRISMA-based review first establishes critical limitations in existing REC modelling approaches, particularly regarding their representation of real-world uncertainties and emergent community-scale phenomena. Subsequent development follows a formal engineering methodology comprising: (1) requirements synthesis from operational REC case studies, (2) architectural enhancement of the simulation framework, and (3) comprehensive validation against both synthetic and empirical datasets. Experimental validation across a 17-building REC Case Study demonstrates framework's capabilities, through the newly developed functionalities: 73.5% reduction in carbon emissions through solar-optimised control strategies, 71% cost reduction in buildings with differentiated optimisation objectives, and maintenance of thermal comfort parameters within ±0.5°C thresholds, all while achieving 18.6% improvement in grid stability metrics. These results provide quantitative evidence for the simultaneous optimisation of sustainability, economic, and reliability objectives in REC operations. The methodological contributions provide immediate value for energy community operators while establishing essential foundations for future Digital Twin development. For policymakers, the empirical results offer verifiable benchmarks for REC design, particularly in balancing environmental targets with operational constraints. By advancing simulation capabilities toward Digital Twin standards, this work provides a crucial missing link between theoretical energy transition models and real-world deployment requirements.
The global transition toward sustainable energy systems necessitates advanced methodological frameworks capable of managing the complex, stochastic dynamics inherent in Renewable Energy Communities (RECs). Current simulation paradigms remain insufficient for capturing the multi-scale interactions between distributed energy resources, prosumer behaviour, and grid constraints. This dissertation presents significant advances to the CityLearn simulation platform, an open-source platform for Multi-Agent Reinforcement Learning (MARL) in building energy management, through five key methodological innovations that collectively advance its fidelity toward Digital Twin standards: (1) temporal granularity adaptation enabling dynamic resolution switching across sub-hourly intervals, (2) expanded asset integration incorporating household appliances including washing machines, (3) multi-objective optimisation reconciling competing sustainability and economic goals, (4) stochastic realism through probabilistic modelling of generation and demand uncertainties, and (5) standardised data export serialization for interoperability. A systematic PRISMA-based review first establishes critical limitations in existing REC modelling approaches, particularly regarding their representation of real-world uncertainties and emergent community-scale phenomena. Subsequent development follows a formal engineering methodology comprising: (1) requirements synthesis from operational REC case studies, (2) architectural enhancement of the simulation framework, and (3) comprehensive validation against both synthetic and empirical datasets. Experimental validation across a 17-building REC Case Study demonstrates framework's capabilities, through the newly developed functionalities: 73.5% reduction in carbon emissions through solar-optimised control strategies, 71% cost reduction in buildings with differentiated optimisation objectives, and maintenance of thermal comfort parameters within ±0.5°C thresholds, all while achieving 18.6% improvement in grid stability metrics. These results provide quantitative evidence for the simultaneous optimisation of sustainability, economic, and reliability objectives in REC operations. The methodological contributions provide immediate value for energy community operators while establishing essential foundations for future Digital Twin development. For policymakers, the empirical results offer verifiable benchmarks for REC design, particularly in balancing environmental targets with operational constraints. By advancing simulation capabilities toward Digital Twin standards, this work provides a crucial missing link between theoretical energy transition models and real-world deployment requirements.
Description
Keywords
Digital Twin Renewable Energy Communities Smart Grids Energy Simulation Temporal Granularity Energy Optimisation Comunidade de energia renovável Redes energéticas inteligentes Simulação de energia granularidade temporal Otimização energética