Self-Organized Robust Optimization in Open Technical Systems: Self-Organization and Computational Trust for Scalable and Robust Resource Allocation under Uncertainty

Resource allocation, in terms of balancing supply and demand, is a common problem in supply systems, such as electric power systems. Given that these systems are mission-critical – that is, their failure can have massive consequences for people, industries, and public services –, it is of the utmost importance that they maintain the balance under all circumstances. If the system components cannot arbitrarily change their supply for the sake of balance within a fixed period of time, resources have to be allocated in the form of schedules for a number of time steps in advance. In future power systems, maintaining the balance between supply and demand will become an extremely challenging optimization task. Such systems will be characterized by a vast number of distributed energy resources, including weather-dependent power plants and small dispatchable generators, as well as new types of consumers. A key aspect to deal with the complexity and the uncertainties in future power systems is to enable the system components to act autonomously in their environment, to maintain efficient organizational structures, and to anticipate uncertainties originating from the behavior of the other components.

The result of this thesis is an integrated approach to robust resource allocation in open technical systems that is based on the principles of self-organization and computational trust. It introduces Trust-Based Scenario Trees as a trust model to quantify and anticipate uncertainties emanating from volatile demand that follows different behavioral patterns. Trust-Based Scenario Trees function as the basis for finding robust solutions to the scheduling problem, that is, the optimization problem of creating suitable schedules. Further, this thesis presents methods for self-organizing hierarchical system structures that serve as an approach to autonomous problem decomposition in large-scale open technical systems. These methods comprise partitioning constraints, homogeneous partitioning as an underlying organizational paradigm, and the two self-organization algorithms PSOPP and SPADA. While the partitioning constraints specify the shape of the hierarchy, homogeneous partitioning defines the desired composition of the subsystems residing in the hierarchy. The two self-organization algorithms PSOPP and SPADA enable the system components to maintain an adequate hierarchical structure that supports the system's goals. These methods lay the foundation for the system's robustness, efficiency, and scalability. Moreover, the thesis outlines concepts and optimization algorithms for robust resource allocation in self-organizing hierarchies. In detail, it specifies robust solutions to the scheduling problem that allow the system components to deal with different possible developments of the demand; created schedules rely on Trust-Based Scenario Trees. For the timely creation of high-quality robust solutions, the thesis presents the auction- and trust-based scheduling algorithm TruCAOS that reduces the complexity of the scheduling problem by enabling the components to actively participate in the process of schedule creation.

All concepts and algorithms devised in this thesis have been analyzed in extensive empirical evaluations in an elaborate simulation environment for autonomous power systems on the basis of real world data.

published 12.07.2017 in: Augsburg Publikationsserver OPUS der Universitätsbibliothek Augsburg


For questions regarding the publication, please contact!