Future Energy Grids and Smart Cities
The climate changes, the catastrophe in Fukushima, the recent biggest blackout in history in India due to an overloaded electricity grid or the dwindling oil reserves world-wide are some of the manifold different reasons why countries massively increase their efforts in shaping their future energy generation, distribution, transportation and consumption, in short in future energy grids (FEG). Although much is still in a state of flux it is nevertheless commonly accepted that existing grids cannot simply be extended to address the requirements of the next generation of energy supply and consumption. Instead, a fundamental re-engineering of the grids is required.
Besides the topic of future energy grids the recent past was also dominated by the discussion about so called smart cities. A smart city uses information and communication technologies (ICT) to enhance quality, performance and interactivity of urban services. This especially means that the contact between citizens and government is eased and improved substantially with the aim to equip inhabitants with more power, responsibility and easing their life substantially from bureaucratic and useless tasks. Another highly relevant goal is to reduce costs and resource consumption. Smart cities will connect, utilize and optimize a number of sectors including transport and traffic management, energy consumption and management or water and waste issues. This, however, implies that smart cities and smart grids need to be deeply integrated in order to get the best of both of them.
Smart Grids and Smart Cities
Agent.HyGrid: A seamless Development Process for Agent-based Control Solutions in Hybrid Energy Infrastructures and Smart Cities
Decisions regarding design, implementation and needed policies for decentralized control solutions of future energy grids require profound investigations that ensure a secure and reliable grid operation. This applies to scientific computational systems serving the development of new control approaches on the one hand, but also for test-bed frameworks being used for the final verification of required functionalities before hard- and software components are deployed in real applications on the other hand. In an ideal case, the developed software artefacts are reused in various on-site systems for different purposes, which would avoid a redundant work overhead. By closing the gap between simulation environments, test-beds and real on-site applications, Agent.HyGrid intends to demonstrate that such systematic and seamless software development process is feasible. Based on the definition and the unifying concept of so called “Energy Agents”, a reference development process is defined. It shall be used as a blueprint for further developments of decentralized, agent-based control solutions. Artificial Intelligent.
Electricity Wholesale Markets & Power to Gas
Governments have started to declare their energy transition policies in order to create a projection for their future energy landscape. Renewable technologies, such as roof-top solar panels or wind turbines, are improving and becoming much more widespread. Therefore, these transition policies will eventually shift energy supply from controllable production to less controllable production. As of 2015, Germany has produced 25-30% of its total energy production from renewable resources (80 GW capacity); on some days already up to 60% of the overall consumption. In 2030, it is planned that the renewable capacity will reach the 125 GW limit. Since the renewable production is highly weather depended, it has already started to significantly affect market prices, leading to highly volatile market regimes and substantial price fluctuations. We conduct several researches to tackle the feasibility of wholesale power to hydrogen systems for future scenarios using a powerful and competitive smart grid simulation platform.
Traditional power infrastructures suffer from unidirectional power flow from bulk generation to consumers. High penetration of distributed energy generation in the low and medium voltage level brings about a major change in the topology of power infrastructures. Having the centralized transmissions and distribution management systems, current technology is not capable of providing all the functionalities needed for the future Smart Grid. One of the possible solutions is to group, manage and supervise distributed local generation and consumption, storage facilities, and loads locally. This will form interconnected networks working autonomously under different system conditions. These are called micro-grids. They operate both, being connected to the main power grid as well as in an isolated mode. This makes the control and reliability issues much more complex than in the traditional centralized systems. In our chair we study various control strategies of micro-grids, utilizing the benefits that multi-agent based systems and simulations provide. The goal is to address the newly arisen issues related to decentralize micro-grid management.
Machine Learning in Electricity Wholesale Markets
The role of local wholesale markets is becoming more and more significant since renewable electricity production is increasing significantly worldwide. This results in difficult to predict production volumes that may cause frequent price fluctuations. Machine learning is able to cope with this problem by means of utilizing statistical, data-driven decision support models. Our chair is implementing a number of trading models to have a closer look at the power-to-hydrogen phenomenon. Proposed approaches are evaluated under two scenarios: Grid balancing and fuel-cell based power management and transportation. Grid balancing has been used by Transmission System Operators (TSO) to bridge, respectively overcome weak points within the grid network. However, it has not been considered on the local level. While still being in its infancy hydrogen based management and transportation is becoming more and more popular due to the fact that it is a reasonable solution to deal with (sudden) excess electricity.
Medical Diagnosis Systems
Learning in Multi-agent-based Medical Diagnosis Systems
The field of Multi-Agent Systems (MAS) aims to provide intelligence as one of its key characteristics. Intelligence and learning are interdependent as learning is the self-improvement of the future behavior based on past experiences. Thus multi-agent systems are expected to be equipped with learning abilities. Multi-Agent Learning is the integration of Machine Learning techniques and multi-agent systems. The work at our chair is focused on the adaptation of learning methods to a Stigmergic Medical Diagnosis System previously designed in our chair based on MAS. Since our system’s functionality in based on Ant Colony Optimization algorithm and its structure is designed according to the holonic paradigm. Project-specific reinforcement learning and clustering methods are now being developed and implemented in order to make the system capable of learning.
Empirical Evaluation of Programming Language and Software Engineering Constructs
While there is a large variety of different constructs available in the field of programming languages and software engineering, it is rather rarely the case that the actual, measurable benefit of such constructs is shown. Our chair works on the empirical evaluation of programming language and software engineering constructs in general, and the evaluation of such constructs using controlled experiments in particular. Most of the controlled experiments in the area of type systems (i.e. the comparison of static and dynamic type systems) world wide was done by our chair. Additionally, we ran in the past studies on for example aspect-oriented language constructs, on modeling notations, or on tools such as code completion. Our general approach on this topic is that we let a number of participants work on a certain programming or design task and measure things such as the time required to solve such tasks or the number of errors that occured. The goal of our studies is not only to apply empirical methods but to investigate what kinds of studies can be practically performed. The overall goal of these studies is to test, what constructs do finally provide measurable benefits - and which ones don't.