| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 6.19 MB | Adobe PDF |
Authors
Abstract(s)
Ao longo das últimas décadas, os mercados de energia elétrica têm sofrido grandes alterações no seu funcionamento de forma a dar resposta aos desafios emergentes. Essa evolução contribuiu para um grande aumento da sua complexidade e imprevisibilidade, dificultando a participação das entidades envolvidas. De forma a possibilitar lidar com a dificuldade identificada, surgiram várias ferramentas para o estudo dos mercados, permitindo a análise das diversas entidades envolvidas, as suas interações assim como as regras em vigor. No entanto, as soluções existentes são principalmente focadas nos modelos de mercado baseados em leilão, apresentando um apoio muito reduzido, e em alguns casos inexistente, ao modelo de mercado baseado na negociação de contratos bilaterais.
Com o intuito de colmatar a necessidade identificada, a presente dissertação propõe o desenvolvimento de um sistema de apoio à decisão multi-agente para a negociação de contratos bilaterais em mercados de energia elétrica. Para esse fim, o sistema oferece apoio à decisão nas fases de pré-negociação e negociação de contratos bilaterais.
Na fase de pré-negociação são identificados os oponentes, assim como a quantidade de energia a transacionar com cada um e o respetivo preço esperado, de forma a potenciar o maior benefício para o player apoiado, tendo em conta os seus objetivos. A recomendação final resulta da reunião de vários processos, incluindo: deteção de contextos de negociação; geração de cenários de negociação alternativos; aprendizagem sobre qual o método de geração de cenários mais realista; análise das ações possíveis que o player pode tomar; gestão de risco, considerando a reputação de cada oponente; e o processo de decisão que permite a adoção de diferentes estratégias para a identificação da ação com maior benefício.
Na fase de negociação é identificada, a cada oferta/contra-oferta, a melhor estratégia que o player apoiado pode seguir com determinado oponente, num determinado contexto, de forma a obter o melhor resultado possível na sua negociação. A recomendação resulta de um processo de aprendizagem, inspirado no funcionamento das redes neuronais artificiais, de forma a confrontar informação de fontes distintas: experiência pessoal e partilhada por outros players acerca do oponente, dos oponentes com um perfil de negociação semelhante e de todos os oponentes no geral. Este processo tem como componentes chave: a deteção do contexto de negociação; identificação de perfis de negociação; atribuição de pesos às diferentes fontes de informação conforme o seu contributo; e gestão de credibilidade de cada player tendo em conta a qualidade da informação partilhada.
Os modelos desenvolvidos são integrados em diferentes agentes, capazes de se adaptarem a diferentes contextos de negociação e comportamentos dos oponentes em questão. Para esse efeito são aplicados conceitos de aprendizagem automática, matemática e negociação automática. A solução proposta é testada e validada através de simulação utilizando cenários baseados em dados reais do mercado ibérico de energia elétrica.
Over the last few decades, electricity markets have undergone major changes in their operation in order to respond to the emerging challenges. This evolution has contributed to a great increase in its complexity and unpredictability, making it difficult for the involved entities to participate. In order to deal with the difficulty identified, several tools emerged for the study of markets, allowing the analysis of the various involved entities, their interactions as well as the applicable rules. However, existing solutions are mainly focused on auction-based market models, with very limited, and in some cases non-existent, support for the market model based on the bilateral contracts negotiation. In order to address the identified need, this dissertation proposes the development of a multi-agent decision support system for the negotiation of bilateral contracts in electricity markets. To this end, the system provides support for the decision in the pre-negotiation and negotiation phases of bilateral contracts. In the pre-negotiation phase, the opponents are identified, as well as the amount of energy to be transacted with each one and its expected price, in order to maximize the benefit to the supported player, taking into account its objectives. The final recommendation results from the reunion of several processes, including: detection of negotiation contexts; generation of alternative negotiation scenarios; learning about which method of scenario generation is the most realistic; analysis of possible actions that player can take; risk management, taking into account the reputation of each opponent; and the decision process that allows the adoption of different strategies for the identification of the action with greater benefit. In the negotiation phase, at each offer/counter-offer, is identified the best strategy that the supported player can follow with a certain opponent, in a given context, in order to obtain the best possible outcome in its negotiation. The recommendation results from a learning process, inspired by the operation of artificial neural networks, in order to confront information from different sources: personal experience shared by other players about the opponent; opponents with a similar negotiating profile; and of all opponents in general. This process has as key components: the detection of the negotiation context; identification of negotiation profiles; attribution of weights to the different sources of information according to their contribution; and credibility management of each player taking into account the quality of the shared information. The developed models are integrated in different agents, able to adapt to different contexts of negotiation and behaviors of the opponents in matter. For this purpose, concepts of machine learning, mathematics and automated negotiation are applied. The proposed solution is tested and validated through simulation, using scenarios based on real data from the Iberian electricity market.
Over the last few decades, electricity markets have undergone major changes in their operation in order to respond to the emerging challenges. This evolution has contributed to a great increase in its complexity and unpredictability, making it difficult for the involved entities to participate. In order to deal with the difficulty identified, several tools emerged for the study of markets, allowing the analysis of the various involved entities, their interactions as well as the applicable rules. However, existing solutions are mainly focused on auction-based market models, with very limited, and in some cases non-existent, support for the market model based on the bilateral contracts negotiation. In order to address the identified need, this dissertation proposes the development of a multi-agent decision support system for the negotiation of bilateral contracts in electricity markets. To this end, the system provides support for the decision in the pre-negotiation and negotiation phases of bilateral contracts. In the pre-negotiation phase, the opponents are identified, as well as the amount of energy to be transacted with each one and its expected price, in order to maximize the benefit to the supported player, taking into account its objectives. The final recommendation results from the reunion of several processes, including: detection of negotiation contexts; generation of alternative negotiation scenarios; learning about which method of scenario generation is the most realistic; analysis of possible actions that player can take; risk management, taking into account the reputation of each opponent; and the decision process that allows the adoption of different strategies for the identification of the action with greater benefit. In the negotiation phase, at each offer/counter-offer, is identified the best strategy that the supported player can follow with a certain opponent, in a given context, in order to obtain the best possible outcome in its negotiation. The recommendation results from a learning process, inspired by the operation of artificial neural networks, in order to confront information from different sources: personal experience shared by other players about the opponent; opponents with a similar negotiating profile; and of all opponents in general. This process has as key components: the detection of the negotiation context; identification of negotiation profiles; attribution of weights to the different sources of information according to their contribution; and credibility management of each player taking into account the quality of the shared information. The developed models are integrated in different agents, able to adapt to different contexts of negotiation and behaviors of the opponents in matter. For this purpose, concepts of machine learning, mathematics and automated negotiation are applied. The proposed solution is tested and validated through simulation, using scenarios based on real data from the Iberian electricity market.
Description
Keywords
Contratos Bilaterais Aprendizagem Automática Mercados de Energia Elétrica Negociação Automática Sistema de Apoio à Decisão Sistema Multi-Agente