No description available.

Wholesale electricity markets now operate in many countries around the world. These markets determine a spot price for electricity as the clearing price when generators bid in energy at various prices. As the trading in a wholesale electricity market can be seen as a dynamic repeated game, it would be expected that profit maximising generators learn to engage in tacit collusion to profitably increase spot market prices. This thesis investigates this tacit collusion of generators in oligopolistic electricity markets. We do not follow the approach of previous work in game theory that presupposes firms' collusive strategies to enforce collusion in an oligopoly. Instead, we develop a co-evolutionary approach (extending previous work in this area) using a genetic algorithm (GA) to co-evolve strategies for all generators in some stylised models of an electricity market. The bidding strategy of each generator is modelled as a set of bidding actions, one for each possible discrete state of the state space observed by the generator. The market trading interactions are simulated to determine the fitness of a particular strategy. The tacitly collusive outcomes and strategies emerging from computational experiments are thus obtained from the learning or evolutionary process instead of from any pre-specification. Analysing many of those emergent collusive outcomes and strategies. we are able to specify the mechanism of tacit collusion and investigate how the market environment can affect it. We find that the learned collusive strategies are similar to the forgiving trigger strategies of classical supergame theory (Green and Porter, 1984). Also using computational experiments, we can determine which characteristics of the market environment encourage or hinder tacit collusion. The findings from this thesis provide insights on tacit collusion in an oligopoly and policy implications from a learning perspective. With modelling flexibility, our co-evolutionary approach can be extended to study strategic behaviour in an oligopoly considering many other market characteristics.

Wholesale electricity markets now operate in many countries around the world. These markets determine a spot price for electricity as the clearing price when generators bid in energy at various prices. As the trading in a wholesale electricity market can be seen as a dynamic repeated game, it would be expected that profit maximising generators learn to engage in tacit collusion to profitably increase spot market prices. This thesis investigates this tacit collusion of generators in oligopolistic electricity markets. We do not follow the approach of previous work in game theory that presupposes firms' collusive strategies to enforce collusion in an oligopoly. Instead, we develop a co-evolutionary approach (extending previous work in this area) using a genetic algorithm (GA) to co-evolve strategies for all generators in some stylised models of an electricity market. The bidding strategy of each generator is modelled as a set of bidding actions, one for each possible discrete state of the state space observed by the generator. The market trading interactions are simulated to determine the fitness of a particular strategy. The tacitly collusive outcomes and strategies emerging from computational experiments are thus obtained from the learning or evolutionary process instead of from any pre-specification. Analysing many of those emergent collusive outcomes and strategies. we are able to specify the mechanism of tacit collusion and investigate how the market environment can affect it. We find that the learned collusive strategies are similar to the forgiving trigger strategies of classical supergame theory (Green and Porter, 1984). Also using computational experiments, we can determine which characteristics of the market environment encourage or hinder tacit collusion. The findings from this thesis provide insights on tacit collusion in an oligopoly and policy implications from a learning perspective. With modelling flexibility, our co-evolutionary approach can be extended to study strategic behaviour in an oligopoly considering many other market characteristics.

Wholesale electricity markets now operate in many countries around the world. These markets determine a spot price for electricity as the clearing price when generators bid in energy at various prices. As the trading in a wholesale electricity market can be seen as a dynamic repeated game, it would be expected that profit maximising generators learn to engage in tacit collusion to profitably increase spot market prices. This thesis investigates this tacit collusion of generators in oligopolistic electricity markets. We do not follow the approach of previous work in game theory that presupposes firms' collusive strategies to enforce collusion in an oligopoly. Instead, we develop a co-evolutionary approach (extending previous work in this area) using a genetic algorithm (GA) to co-evolve strategies for all generators in some stylised models of an electricity market. The bidding strategy of each generator is modelled as a set of bidding actions, one for each possible discrete state of the state space observed by the generator. The market trading interactions are simulated to determine the fitness of a particular strategy. The tacitly collusive outcomes and strategies emerging from computational experiments are thus obtained from the learning or evolutionary process instead of from any pre-specification. Analysing many of those emergent collusive outcomes and strategies. we are able to specify the mechanism of tacit collusion and investigate how the market environment can affect it. We find that the learned collusive strategies are similar to the forgiving trigger strategies of classical supergame theory (Green and Porter, 1984). Also using computational experiments, we can determine which characteristics of the market environment encourage or hinder tacit collusion. The findings from this thesis provide insights on tacit collusion in an oligopoly and policy implications from a learning perspective. With modelling flexibility, our co-evolutionary approach can be extended to study strategic behaviour in an oligopoly considering many other market characteristics.

Wholesale electricity markets now operate in many countries around the world. These markets determine a spot price for electricity as the clearing price when generators bid in energy at various prices. As the trading in a wholesale electricity market can be seen as a dynamic repeated game, it would be expected that profit maximising generators learn to engage in tacit collusion to profitably increase spot market prices. This thesis investigates this tacit collusion of generators in oligopolistic electricity markets. We do not follow the approach of previous work in game theory that presupposes firms' collusive strategies to enforce collusion in an oligopoly. Instead, we develop a co-evolutionary approach (extending previous work in this area) using a genetic algorithm (GA) to co-evolve strategies for all generators in some stylised models of an electricity market. The bidding strategy of each generator is modelled as a set of bidding actions, one for each possible discrete state of the state space observed by the generator. The market trading interactions are simulated to determine the fitness of a particular strategy. The tacitly collusive outcomes and strategies emerging from computational experiments are thus obtained from the learning or evolutionary process instead of from any pre-specification. Analysing many of those emergent collusive outcomes and strategies. we are able to specify the mechanism of tacit collusion and investigate how the market environment can affect it. We find that the learned collusive strategies are similar to the forgiving trigger strategies of classical supergame theory (Green and Porter, 1984). Also using computational experiments, we can determine which characteristics of the market environment encourage or hinder tacit collusion. The findings from this thesis provide insights on tacit collusion in an oligopoly and policy implications from a learning perspective. With modelling flexibility, our co-evolutionary approach can be extended to study strategic behaviour in an oligopoly considering many other market characteristics.

Wholesale electricity markets now operate in many countries around the world. These markets determine a spot price for electricity as the clearing price when generators bid in energy at various prices. As the trading in a wholesale electricity market can be seen as a dynamic repeated game, it would be expected that profit maximising generators learn to engage in tacit collusion to profitably increase spot market prices. This thesis investigates this tacit collusion of generators in oligopolistic electricity markets. We do not follow the approach of previous work in game theory that presupposes firms' collusive strategies to enforce collusion in an oligopoly. Instead, we develop a co-evolutionary approach (extending previous work in this area) using a genetic algorithm (GA) to co-evolve strategies for all generators in some stylised models of an electricity market. The bidding strategy of each generator is modelled as a set of bidding actions, one for each possible discrete state of the state space observed by the generator. The market trading interactions are simulated to determine the fitness of a particular strategy. The tacitly collusive outcomes and strategies emerging from computational experiments are thus obtained from the learning or evolutionary process instead of from any pre-specification. Analysing many of those emergent collusive outcomes and strategies. we are able to specify the mechanism of tacit collusion and investigate how the market environment can affect it. We find that the learned collusive strategies are similar to the forgiving trigger strategies of classical supergame theory (Green and Porter, 1984). Also using computational experiments, we can determine which characteristics of the market environment encourage or hinder tacit collusion. The findings from this thesis provide insights on tacit collusion in an oligopoly and policy implications from a learning perspective. With modelling flexibility, our co-evolutionary approach can be extended to study strategic behaviour in an oligopoly considering many other market characteristics.

Wholesale electricity markets now operate in many countries around the world. These markets determine a spot price for electricity as the clearing price when generators bid in energy at various prices. As the trading in a wholesale electricity market can be seen as a dynamic repeated game, it would be expected that profit maximising generators learn to engage in tacit collusion to profitably increase spot market prices. This thesis investigates this tacit collusion of generators in oligopolistic electricity markets. We do not follow the approach of previous work in game theory that presupposes firms' collusive strategies to enforce collusion in an oligopoly. Instead, we develop a co-evolutionary approach (extending previous work in this area) using a genetic algorithm (GA) to co-evolve strategies for all generators in some stylised models of an electricity market. The bidding strategy of each generator is modelled as a set of bidding actions, one for each possible discrete state of the state space observed by the generator. The market trading interactions are simulated to determine the fitness of a particular strategy. The tacitly collusive outcomes and strategies emerging from computational experiments are thus obtained from the learning or evolutionary process instead of from any pre-specification. Analysing many of those emergent collusive outcomes and strategies. we are able to specify the mechanism of tacit collusion and investigate how the market environment can affect it. We find that the learned collusive strategies are similar to the forgiving trigger strategies of classical supergame theory (Green and Porter, 1984). Also using computational experiments, we can determine which characteristics of the market environment encourage or hinder tacit collusion. The findings from this thesis provide insights on tacit collusion in an oligopoly and policy implications from a learning perspective. With modelling flexibility, our co-evolutionary approach can be extended to study strategic behaviour in an oligopoly considering many other market characteristics.

Wholesale electricity markets now operate in many countries around the world. These markets determine a spot price for electricity as the clearing price when generators bid in energy at various prices. As the trading in a wholesale electricity market can be seen as a dynamic repeated game, it would be expected that profit maximising generators learn to engage in tacit collusion to profitably increase spot market prices. This thesis investigates this tacit collusion of generators in oligopolistic electricity markets. We do not follow the approach of previous work in game theory that presupposes firms' collusive strategies to enforce collusion in an oligopoly. Instead, we develop a co-evolutionary approach (extending previous work in this area) using a genetic algorithm (GA) to co-evolve strategies for all generators in some stylised models of an electricity market. The bidding strategy of each generator is modelled as a set of bidding actions, one for each possible discrete state of the state space observed by the generator. The market trading interactions are simulated to determine the fitness of a particular strategy. The tacitly collusive outcomes and strategies emerging from computational experiments are thus obtained from the learning or evolutionary process instead of from any pre-specification. Analysing many of those emergent collusive outcomes and strategies. we are able to specify the mechanism of tacit collusion and investigate how the market environment can affect it. We find that the learned collusive strategies are similar to the forgiving trigger strategies of classical supergame theory (Green and Porter, 1984). Also using computational experiments, we can determine which characteristics of the market environment encourage or hinder tacit collusion. The findings from this thesis provide insights on tacit collusion in an oligopoly and policy implications from a learning perspective. With modelling flexibility, our co-evolutionary approach can be extended to study strategic behaviour in an oligopoly considering many other market characteristics.

CNPq / Nesta tese, os Sistemas Imunológicos Artificiais são aplicados a diferentes instâncias do despacho econômico e econômico/ambiental de energia elétrica. Os sistemas imunes considerados são baseados no princípio da seleção clonal e usam uma representação real com operador de \emph{aging} puro e operadores de hipermutação utilizando distribuições de probabilidade Gaussianas e de Cauchy. Algoritmos Culturais utilizando fontes de conhecimento normativo, situacional, histórico e topográfico são incorporados para melhorar a capacidade de otimização global dos sistemas imunes. Todas as abordagens propostas possuem vários pontos de auto-adaptação e a maioria utiliza um operador de busca local baseado na técnica quase-simplex. Uma sequência caótica também é considerada como uma potencial fonte de melhoria na variação cultural do algoritmo. Procedimentos de reparação constituem outra contribuição do trabalho e são aplicados para evitar lidar com soluções infactíveis em todos os problemas abordados. Na primeira parte dos experimentos, quatro instâncias do problema do despacho econômico de energia são consideradas. Em todos os casos, foi utilizada uma função não suave de custo de combustível levando em consideração os efeitos de ponto de válvula. Uma das instâncias também considera as perdas na transmissão de energia. Nos experimentos de comparação entre as abordagens propostas, as versões imuno-culturais superam a versão puramente imune. O método cultural proposto que apresenta melhor resultado é escolhido para ser comparado a outras técnicas modernas de otimização reportadas na literatura recente. Em todos os casos mono-objetivo considerados, a abordagem proposta é capaz de encontrar o menor custo de combustível. A segunda parte dos experimentos trata do problema do despacho econômico/ambiental. Esta é uma versão do problema do despacho econômico de energia onde a emissão de poluentes é adicionada como um novo objetivo, tornando este um problema de otimização multiobjetivo não-linear com restrições. Algoritmos imuno-culturais baseados em fatores de escalarização e dominância de Pareto são propostos para este caso. Várias instâncias do problema são utilizadas nos experimentos, algumas das quais consideram perdas na transmissão de energia. Os algoritmos propostos são favoravelmente comparados com um algoritmo do estado-da-arte para otimização multiobjetivo. O melhor algoritmo proposto também é comparado com métodos reportados na literatura recente. As comparações mostram o bom desempenho da melhor abordagem proposta e confirmam seu potencial para resolver o problema do despacho econômico/ambiental de energia. / In this thesis, Artificial Immune Systems are applied to solve different instances of the economic and environmental/economic load dispatch problems. The immune systems considered here are based on the clonal selection principle and use a real coded representation with pure aging operator and hypermutation operators utilizing Gaussian and Cauchy distributions. Cultural Algorithms using normative, situational, historical and topographical knowledge sources are incorporated to improve the global optimization capability of immune systems. All the proposed approaches have several points of self-adaptation and most of them use a local search operator that is based on a quasi-simplex technique. A chaotic sequence is also considered as a potential source of improvement to the cultural variation. Repair procedures represent another contribution of this work and are applied to avoid dealing with infeasible solutions in all the considered problems. In the first part of the experiments, four instances of the economic load dispatch problem are considered. In all the cases, a non-smooth fuel cost function which takes into account the valve-point loading effects is utilized. One of instances also considers energy transmission losses. In the experiments conducted to compare the proposed approaches, the immune-cultural based approaches outperformed the pure immune version. The proposed cultural method which presents the best performance is chosen to be compared with other modern optimization techniques reported in the recent literature. In all the mono-objective cases considered, the proposed approach is capable of finding the minimum fuel cost value. The second part of the experiments deals with the environmental/economic load dispatch problem. This is a multi-objective version of the economic load dispatch where pollution emission is added as an objective, it is formulated as a non-linear constrained multi-objective optimization problem. Cultural immune algorithms based on scalarizing factors and Pareto-dominance are proposed for this case. Several instances of the problem are considered, some dealing with energy transmission losses. The proposed algorithms are favorably compared with a state-of-art algorithm for multi-objective optimization (the Non-dominated Sorting Genetic Algorithm II - NSGA - II). The best proposed algorithm is also compared with methods reported in recent literature. The comparisons demonstrate the good performance of the best proposed approach and confirm its potential to solve the environmental/economic load dispatch problem.