Start Date: October 1, 2014
Akasiadis Charilaos
Balancing energy demand and production in modern Smart Grids with increased penetration of intermittent renewable energy resources is a challenging problem. Demand-Side Management (DSM), i.e., the design and application of sophisticated mechanisms for managing and coordinating energy demand, has been hailed as a means to deal with this problem. In this dissertation, we propose mechanisms for the formation of agent cooperatives offering large-scale DSM services, and put forward a complete framework for their operation. Individuals, being either mere consumers, or even prosumers of electricity, are represented by rational agents and form coalitions to offer demand shifting from peak to non-peak intervals.For cooperatives of consumers, we present an effective consumption shifting scheme, equipped with desirable guarantees, such as individual rationality, truth-fulness, and (weak) budget balance. Our scheme employs several algorithms to promote the formation of the most effective shifting coalitions. It takes into account the shifting costs of the individuals, and rewards them according to their shifting efficiency. In addition, it employs internal pricing methods that guarantee individ-ual rationality, and allow agents with initially forbidding costs to also contribute to the shifting effort. The truthfulness of agent statements regarding their shifting behaviour is ascertained via the incorporation of a strictly proper scoring rule. We provide a thorough evaluation of our approach on a simulations setting constructed over a real-world dataset. Our simulation results clearly demonstrate the benefits arising from the use of agent cooperatives in this domain.
Moreover, to also allow the decentralized coordination of cooperatives of prosumers, we combine, for the first time in the literature, a strictly proper scoring rule with a specialized cryptocurrency framework. Using our approach, prosumers col-laborate with the use of a blockchain-oriented framework to manage their demand, in order to make more profits from the selling of their energy. When tested on a simulation setting that uses dynamic electricity pricing to promote the usage of lo-cally generated renewable energy, our approach drives the prosumers to become more engaged in DSM and achieve increased profits; the balancing of demand and local renewable supply is more effective; and dynamic electricity prices are more stable.
Furthermore, we propose a vehicle-to-grid/grid-to-vehicle (V2G/G2V) algorithm that balances demand and local renewable supply in environments populated with electric vehicles. The approach promotes new business models that make effective use of the capability of electric vehicles to store energy in their batteries. Additionally, to assess participating agents’ uncertainty, and correctly predict their future behaviour regarding power consumption shifting actions, promoting in this way accuracy and effectiveness, we adopt various machine learning techniques, adapt them to fit the problem domain, and use these to effectively monitor the trustworthiness of agent statements regarding their final shifting actions. Simulation results confirm that the adoption of machine learning techniques provides tangible benefits regarding enhanced cooperative performance, and increased financial gains for the participants.
Finally, we provide the methodology for delivering large-scale DSM services in the real world. To this purpose, we devise an IoT service-oriented architecture for DSM applications, through which we test different GUIs and incentive types for managing energy consumption. In this context, we present a “serious game” solution that was tested by real human subjects. Our approach comes complete with the adoption of a statistical analysis methodology to validate reductions in consumption and the promotion of renewable energy usage in real world settings. Our results show that using the proposed methods in real-world large-scale settings can significantly benefit the end-users, the Grid, and the environment. The success of our approach indicates that the combination of methods from multiple fields of Computer Science can deliver high quality human-centered solutions to complex real-world problems.