Jörg Bremer

Encoding distributed search spaces for virtual power plants

The optimization task in many virtual power plant (VPP) scenarios comprises the search for appropriate schedules in search spaces from distributed energy resources. In scenarios with a decoupling of plant modeling and plant control, these search spaces are distributed as well. If merely the controller unit of a plant knows about the subset of operable schedules that are allowed to be considered by the central scheduling unit, then these sets have to be effectively communicated. We discuss an approach of learning the envelope that separates operable from non-operable schedules inside the space of all schedules by means of support vector data description. Then, only the comparatively small set of support vectors has to be transmitted as a classifier for distinguishing schedules during optimization. We applied this approach to simulated VPP.

Distributed hybrid constraint handling in large scale virtual power plants

In many virtual power plant (VPP) scenarios, numerous individually configured units within a VPP have to be scheduled regarding both global constraints (i.e. external market demands) and local constraints (i.e. technical, economical or ecological aspects for each unit). Approaches for global and local constraint handling have been discussed in the relevant literature independently. A hybrid approach is proposed that combines a decentralized combinatorial optimization heuristic with the encoding of individually constrained search spaces into unconstrained representations by means of support vector data description. The approach is applied to simulated VPP.

Support vector based encoding of distributed energy resources' feasible load spaces

The sets of feasible load schedules that distributed energy resources are able to operate, jointly define the search space within many virtual power plant optimization tasks. If a centralized approach is considered, a central, single scheduling unit needs to know for each energy resource what schedules comply with all given constraints, because only these are operable and might be taken into account for optimization. As many constraints depend on state or time, sets of currently operable alternatives have repeatedly to be communicated to the scheduler in order to avoid central modeling of each single resource. We here present a support vector based approach for learning a highly efficient geometric representation of the space of feasible alternatives for operable schedules. This description is communicated to the scheduler and the encoded information implicitly contains all constraints and therefore makes their modeling dispensable at scheduler side.

Conjoint dynamic aggregation and scheduling methods for dynamic virtual power plants

The increasing pervasion of information and communication technology (ICT) in energy systems allows for the development of new control concepts on all voltage levels. In the distribution grid, this development is accompanied by a still increasing penetration with distributed energy resources like photovoltaic (PV) plants, wind turbines or small scale combined heat and power (CHP) plants. Combined with shiftable loads and electrical storage, these energy units set up a new flexibility potential in the distribution grid that can be tapped with ICT-based control following the long-term goal of substituting conventional power generation. In this contribution, we propose an architectural model and algorithms for the self-organization of these distributed energy units within dynamic virtual power plants (DVPP) along with first results from a feasibility study of the integrated process chain from market-driven DVPP formation to product delivery.

Parallel tempering for constrained many criteria optimization in dynamic virtual power plants

Following the long-term goal of substituting conventional power generation, market oriented approaches will lead to interaction, competition but also collaboration between different units. Together with the expected huge number of actors, this in turn will lead to a need for self-organized and distributed control structures. Virtual power plants are an established idea for organizing distributed generation. A frequently arising task is solving the scheduling problem that assigns an operation schedule to each energy resource taking into account a bunch of objectives like accurate resemblance of the desired load profile, robustness of the schedule, costs, maximizing remaining flexibility for subsequent planning periods, and more. Nevertheless, also such dynamic approaches exhibit sub-problems demanding for centralized solutions for ahead of time scheduling of active power. In this paper we develop a hybrid approach combining the advantages of parallel tempering with a constraint handling technique based on a support vector decoder for systematically generating solutions; thus ensuring feasible overall solutions. We demonstrate the applicability with a set of simulation results comprising many objective scheduling for different groups of energy resources.

Decentralized Coalition Formation in Agent-Based Smart Grid Applications

A steadily growing pervasion of the energy grid with communication technology is widely seen as an enabler for new computational coordination techniques for renewable, distributed generation as well as for controllable consumers. One important task is the ability to group together in order to jointly gain enough suitable flexibility and capacity to assume responsibility for a specific control task in the grid. We present a fully decentralized coalition formation approach based on an established heuristic for predictive scheduling with the additional advantage of keeping all information about local decision base and local operational constraints private. The approach is evaluated in several simulation scenarios with different type of established models for integrating distributed energy resources.

Estimating Shapley Values for Fair Profit Distribution in Power Planning Smart Grid Coalitions

In future, highly dynamic energy grids a likely scenario is to have dynamically founded groups of distributed energy resources that are in charge of jointly delivering a demanded load schedule for a certain time horizon. In market based scenarios, such a demanded load schedule would be a (day ahead) product that is to be delivered by a coalition of energy resources. Computational aspects of the underlying optimization problem or of proper coalition formation are already subject to many research efforts. In this paper, we focus on the question of fairly sharing the profit among the members of such a coalition. Distributing the surplus merely based on the absolute (load) contribution does not take into account that smaller units maybe provide the means for fine grained control as they are able to modify their load on a smaller scale. Shapley values provide a concept for the decision on how the generated total surplus of an agent coalition should be spread. In this paper, we propose a scheme for efficiently estimating computationally intractable Shapley values as a prospective base for future surplus distribution schemes for smart grid coalitions and discuss some first ideas on how to use them for smart grid active power product coalitions.

Hybrid Multi-ensemble Scheduling

A steadily increasing pervasion of the electrical distribution grid with rather small renewable energy resources imposes fluctuating and hardly predictable feed-in, a partly reverse load flow and demands new predictive load planning strategies. For predictive scheduling with high penetration of renewable energy resources, agent-based approaches using classifier-based decoders for modeling individual flexibilities have shown good performance. On the other hand, such decoder-based methods are currently designed for single entities and not able to cope with ensembles of energy resources. Combining training sets sampled from individually modeled energy units, results in folded distributions with unfavorable properties for training a decoder. Nevertheless, this happens to be a quite frequent use case, e. g. when a hotel, a small business, a school or similar with an ensemble of co-generation, heat pump, solar power, and controllable consumers wants to take part in decentralized predictive scheduling. In this paper, we propose an extension to an established agent approach for scheduling individual single energy units by extending the agents’ decision routine with a covariance matrix adaption evolution strategy that is hybridized with decoders. In this way, locally managed ensembles of energy units can be included. We show the applicability of our approach by conducting several simulation studies.

Generalized cascade classification model with customized transformation based ensembles

Classification of high-dimensional data with imbalanced classes poses problems. Especially such time series classification tasks are problematic, because the ordering of each time step (feature) is important and therefore dimensionality reduction and feature selection cannot be applied. The cascade classification model was developed for such time series classification tasks. The cascade classifier splits high-dimensional classification tasks into a cascade of low-dimensional tasks. But the cascade classification model can only handle data sets with a data structure that can be easily learned in low-dimensional space. In this paper, we propose a generalized version of the cascade classification model that can also deal with data sets with more complex data structures. Generalization is achieved with time series transformations and an ensemble of classifiers based on the time series classifier: transformation based ensembles. For this purpose the cascade classifier is integrated into transformation based ensembles with some adjustments. In a simulation study we apply the generalized cascade classification model to predict the realizability (feasibility) of power production time series for pools of different numbers of micro combined heat and power plants. We show that the choice of the aggregation scheme for the ensemble members in the generalized cascade classification model has a strong impact on the overall classification results. But the choice of a weighting scheme showed hardly any influences on the classification result. Furthermore, data sets of different complexity (different structures in data space) yielded very similar classification results.

Local soft constraints in distributed energy scheduling

In this contribution we present an approach on how to include local soft constraints in the fully distributed algorithm COHDA for the task of energy units scheduling in virtual power plants (VPP). We show how a flexibility representation based on surrogate models is extended and trained using soft constraints like avoiding frequent cold starts of combined heat and power plants. During the task of energy scheduling, the agents representing these machines include indicators in their choice for a new operation schedule. Using an example VPP we show that our approach enables the agents to reflect local soft constraints without sacrificing the global result quality.