Sistema de gerenciamento de energia para microrredes usando multiagentes e otimização meta-heurística distribuída em ambiente de co-simulação

• Autor(a) principal: Almada, Janaína Barbosa
• Data de Publicação: 2025
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Repositório Institucional da Universidade Federal do Ceará (UFC)
• Texto Completo: http://repositorio.ufc.br/handle/riufc/79913
• topic: CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA; Metaheurísticas; Sistema multiagente; Simulação (Computadores); Microrredes (redes elétricas inteligentes); Otimização distribuída; Metaheuristics; Multiagent systems; Computer simulation; Microgrids (Smart power grids); Distributed optimization

The diversity of energy resources in distribution networks requires new strategies for planning and operation. Microgrids are solutions to integrate renewable sources, energy storage, and demand response, decentralizing operation and using digital technologies for more granular energy markets. The increasing complexity demands alternatives to centralized techniques, especially for quick decisions such as very short-term load dispatch. This work proposes a distributed optimal dispatch strategy for microgrids with multiple energy resources, focusing on scalability and the use of distributed computational resources. The simulation is carried out by modeling agents on the PADE (Python Agent DEvelopment) platform, implementing the parallelism of metaheuristics, and considering distributed hardware with communication between agents. A co-simulation environment coordinated by Mosaik synchronizes information exchange, while a plug-and-play system allows for dynamic modification of the agents. The PSO (Particle Swarm Optimization) and MAPSO (Multi-Agent Particle Swarm Optimization) algorithms are the chosen metaheuristics. For comparison, the algorithms were implemented without parallelism, centralizing stages in a single agent, and simulating parallelism through a multi-agent system. Case studies show that the distributed MAPSO performs better, with lower objective function values and lower relative standard deviation (15.6%), while the centralized PSO had the highest standard deviation (106.9%). Although the resolution time of the distributed MAPSO is up to 3 times longer (average execution time of 9 seconds), this interval is compatible with dispatch performed every 5 minutes. The distributed processing capability and communication technologies make the method viable for practical application. The main contributions include the development of a plug-and-play system for very short-term optimization and the distribution of metaheuristics among agents, increasing fault tolerance. Keywords: