Managing complex Cyber-Physical Energy Systems (CPES) requires solving various optimization problems with multiple objectives and constraints. As distributed control architectures are becoming more popular in CPES for certain tasks due to their flexibility, robustness, and privacy protection, multi-objective optimization must also be distributed. For this purpose, we present MO-COHDA, a fully distributed, agent-based algorithm, for solving multi-objective optimization problems of CPES. MO-COHDA allows an easy and flexible adaptation to different use cases and integration of custom functionality. To evaluate the effectiveness of MO-COHDA, we compare it to a central NSGA-2 algorithm using multi-objective benchmark functions from the ZDT problem suite. The results show that MO-COHDA can approximate the reference front of the benchmark problems well and is suitable for solving multi-objective optimization problems. In addition, an example use case of scheduling a group of generation units while optimizing three different objectives was evaluated to show how MO-COHDA can be easily applied to real-world optimization problems in CPES.

The increasing integration of renewable energy resources requires so-called smart grid services for monitoring, control and automation tasks. To develop innovative solutions and algorithms, simulation environments are used for evaluation. Especially in smart energy systems, we face a variety of heterogeneous simulators representing, e.g., power grids, analysis or control components. The co-simulation framework mosaik can be used to orchestrate the data exchange and time synchronization between individual simulators. So far, the underlying communication infrastructure has often been assumed to be optimal, so that the influence of e.g., communication delays has been neglected. This paper presents the first results of the project cosima, which aims at connecting the communication simulator OMNeT++ to the co-simulation framework mosaik to analyze the resilience and robustness of smart grid services, e.g., multi-agent-based services with respect to simulation performance, scalability, extensibility and usability. This facilitates simulations with realistic communication technologies (such as 5G) and the analysis of dynamic communication characteristics occuring by simulating multiple messages. We could show, how the simulation performance of this coupling improves by using the new discrete event scheduling of mosaik and how the simulation behaves in scenarios with up to 50 agents.

Modern algorithms in the domain of Deep Reinforcement Learning (DRL) demonstrated remarkable successes; most widely known are those in game-based scenarios, from ATARI video games to Go and the StarCraft~\textsc{II} real-time strategy game. However, applications in the domain of modern Cyber-Physical Systems (CPS) that take advantage a vast variety of DRL algorithms are few. We assume that the benefits would be considerable: Modern CPS have become increasingly complex and evolved beyond traditional methods of modelling and analysis. At the same time, these CPS are confronted with an increasing amount of stochastic inputs, from volatile energy sources in power grids to broad user participation stemming from markets. Approaches of system modelling that use techniques from the domain of Artificial Intelligence (AI) do not focus on analysis and operation. In this paper, we describe the concept of Adversarial Resilience Learning (ARL) that formulates a new approach to complex environment checking and resilient operation: It defines two agent classes, attacker and defender agents. The quintessence of ARL lies in both agents exploring the system and training each other without any domain knowledge. Here, we introduce the ARL software architecture that allows to use a wide range of model-free as well as model-based DRL-based algorithms, and document results of concrete experiment runs on a complex power grid.