Content uploaded by Johannes Hiry
Author content
All content in this area was uploaded by Johannes Hiry on Nov 23, 2017
Content may be subject to copyright.
ISSN 2515-0855
doi: 10.1049/oap-cired.2017.0061
www.ietdl.org
Holistic network planning approach:
enhancement of the grid expansion using
the flexibility of network participants
Lars Jendernalik1✉, Dominique Giavarra1, Christoph Engels2,
Johannes Hiry3, Chris Kittl3, Christian Rehtanz3
1
Westnetz GmbH, Distribution Grid North, Dortmund, Germany
2
University of Applied Sciences Dortmund, Faculty of Computer Science, Dortmund, Germany
3
TU Dortmund University, Institute of Energy Systems, Energy Efficiency and Energy Economics,
Dortmund, Germany
✉E-mail: lars.jendernalik@westnetz.de
Abstract: Distribution systems operator (DSO’s) are facing significant challenges in network planning due to the integration
of distributed energy resources (DER), smart grid technologies, E-mobility, regulation and volatile market conditions. In the
previous work, it was shown (a) how DSO’s are able to optimise planning of network assets in the presence of high
uncertainty and (b) how to describe the interdependencies between all market participants in an agent-based fashion to
derive forecasts of prices, generation and demand. Here, the integration of these solutions across all voltage levels by
means of long-term network planning is presented. Thereby, smart grid components extend the traditional grid planning
process. This extension merges time-series-based network planning approaches with models of active and reactive
innovative network participants. This results in robust and future-proofed network planning.
1 Introduction
While there have been several publications in the past considering
the optimal combination of asset strategy and network planning for
upper voltage levels under uncertain conditions [1,2], as well as
different ways of generating time series for grid planning [3], the
presented approach creates added value by merging these two
fields to improve the electricity grid planning process.
The previous work proposed an optimised planning procedure under
consideration of assets’condition [1]. It provides functional valid grid
structures in terms of asset condition and asset capacity (induced by the
load and feeder) by means of replacement, enforcement and expansion
of physical network elements. In contrast, the dynamics of market
participants also allow the utilisation of flexibilities in terms of
demand-side management, adaptation of regulation frameworks,
storage technologies and market mechanisms. Earlier work lead to
an agent-based simulation framework [3] capable of providing
time-series-based grid utilisation analysis. The combination of both
approaches in the scope of ‘Agent.GridPlan’project allows for an
integrated assessment of structural and operational measures and
thus will result in more realistic and optimal solutions for holistic
target grid planning.
2 Approach
This contribution is integrating the previous approaches across all
voltage levels in the following steps:
(i) Input: The provision of global market conditions is
complemented by regulatory rules. The expected energy mixture in
terms of installed capacity is projected into the future. Weather
and supply tasks describe a set of actual load and feeder situation
in a specific hour which have to be applied to the network system.
(ii) Analysis: The usage of the actual network in terms of capacity,
condition (age) and behaviour of market participants is analysed by
means of agent-based-, asset- and power-flow simulations across all
voltage levels using the input-defined previously. The analysis leads
to potential violations of the network capacities or asset conditions in
a so-called hotspot analysis.
(iii) Fitness restoration: An optimal network adaptation is proposed
by optimisation techniques to balance the resulting future grid
between the options of curtailment or structural change [e.g. the
addition of (innovative) assets].
(iv) Systems results: As an outcome, the network planner is able to
prioritise actions due to the probabilities of scenarios. Furthermore,
he is able to compare the real-network behaviour with the
forecasted situations.
3 Results
The integration of the approaches has led to first results: the resulting
planning and forecasting system across the voltage levels is shown in
Fig. 1.
A risk matrix (right-upper box) serves as a decision base for the
prioritisation of actions regarding the transformer between the
high-voltage (HV) and medium-voltage (MV) levels (black arrows):
actions in the lower-left quadrant must be executed and actions in
the upper-right quadrant are unlikely to be applied. The necessary
grid condition information is delivered by the underlying
agent-based, distributed power-flow calculation which provides
technical information, e.g. nodal voltage (right-lower box), as well
as information about the smart market and smart grid interactions.
3.1 Findings on system’s state of the art
The analysis stage (step 2) is based on the current state of the
agent-based simulation tool for an optimal grid extension planning
(SIMONA) developed among others by Technical University
Dortmund. As described by Kays et al. [3], the current state of
SIMONA focuses on the MV and low-voltage levels.
The agent-based programming paradigm is designed to divide a
large and complex problem (such as a detailed voltage-level
integrated time-series-based distribution grid simulation) into
24th International Conference & Exhibition on Electricity Distribution (CIRED)
12-15 June 2017
Session 5: Planning of power distribution systems
CIRED, Open Access Proc. J., 2017, Vol. 2017, Iss. 1, pp. 2312–2315
2312 This is an open access article published by the IET under the Creative Commons
Attribution License (http://creativecommons.org/licenses/by/3.0/)
smaller problems, which can be modelled very easy. In particular, it
aims to represent individual goals of distinct individual entities in an
environment, where the entities are able to alter their own goals and
behaviours during a negotiation process with other individuals [4].
One example could be the low-level modelling of every single
distributed generation (DG) plant. As the simulation framework’s
ability evolves to a voltage-level integrated time-series-based
simulation from HV to low-voltage level, the number of
participants to be modelled, and therefore the computational effort
raises. Already in this early stage of the ‘Agent.GridPlan’project it
can be noted, that, due to the use of the java agent developing
framework JAVA agent development framework (JADE) [5], there
is a certain competition between modelling simplicity and
computational efficiency. SIMONA is facing similar challenges
considering the performance and scalability as presented in [6].
This challenge has to be addressed by innovative solutions to
compute a large system in an acceptable time and with an
acceptable computational effort.
One possibility to reduce the computational effort and in
consequence accelerate the simulation is identified in the way of
modelling DG plants. A simplified extract of SIMONA’soverall
system concept is shown in Fig. 2. The DG agent communicates with
a time agent, a market agent and a weather agent (all not depicted in
Fig. 2). The time agent announces the beginning of a time step and
waits for the message of every grid agent, to ensure the load flow
calculation of the given time step is finished. At the beginning of
every time step, the market agent informs all other agents about the
current wholesale market price (spot market), whereas the weather
agents broadcast information about the most relevant weather
measures (solar irradiance, wind velocity and direction as well as
temperature). All the described communication steps are performed
by sending or receiving JADE messages. For further information on
the overall system concept, please see [1,7,8].
The available level of detail of environmental data, among others,
justifies that every single DG plant inherits the geo position from its
point of common coupling. In consequence, every DG plant (e.g.
wind, photovoltaic panels etc.) connected to the same node has the
same geo location, and therefore receives the same weather
information. If every DG would be modelled as a distinct agent,
the weather agent would send several messages containing the
same information. In other words: no information surplus with
more communicational effort. The increased message traffic would
have a large impact on the run time, if the SIMONA tool should
be executed in parallel on distinct machines. The interexchange
over an added physical network would slow the overall
performance. Even if the tool is running on a single machine, the
same effects as presented in [6] are observable.
Fig. 1 Scope of the integrated planning solution
Fig. 2 Extract of SIMONA’s overall system concept
CIRED, Open Access Proc. J., 2017, Vol. 2017, Iss. 1, pp. 2312–2315
2313This is an open access article published by the IET under the Creative Commons
Attribution License (http://creativecommons.org/licenses/by/3.0/)
To improve performance, it is proposed to create only one DG agent
per node that aggregates the functionality of several DG agents at the
specific node. Hence, the new DG agent receives time, weather and
market information. Following, it provides the sum of generated
power to the environment. A new feature of this agent concept is a
combination of intrinsic mathematical models, representing different
generation technologies [e.g. photovoltaic and combined heat and
power system (CHP) plants], in a single agent implementation. With
this aggregation of different plants and plant technologies, their
potential individual market driven behaviour is not aggregated. Only
the interface to their environment is bundled. This quantity
reduction of agents, and therefore the communicational effort,
enables a distributed execution of the multi-agent simulation (MAS)
on distinct machines. Moreover, it reduces the memory and central
processing unit usage on every client of the computation cluster,
and therefore accelerates the simulation time even if a simulation is
running on a single machine.
A second potential for performance enhancement isidentified in the
coupling between network nodes and network participants. The
SIMONA simulation environment models every network node as a
distinct agent, whose main task is the determination of residual
nodal powers for a distributed load flow calculation by negotiations
with the connected load and DG agents. It receives messages of the
DG and load agents containing information about the current power
infeed or consumption and provides the latest nodal voltage
magnitude, which could be used, e.g. to adjust the reactive power
infeed in dependency of the given nodal voltage (compare [9]).
With raising voltage levels, the nominal apparent power of DGs
does also grow. Additionally special loads, such as industrial
consumers, often have specific requirements to voltage quality and
reliability. However, for grids at MV level or higher it might
happen, that such significant consumers have a distinct point of
common coupling such as a wave energy converter (WEC) farm
being connected to the grid via its own secondary substation. From
SIMONA’s modelling point of view, this means that these node
agents only need to communicate to a single load or DG agent. In
that case, there is no advantage in splitting up the different agent’s
functionalities. Here, again a new aggregation stage is introduced.
The SIMONA user is given the possibility to choose up to which
number of loads and DGs connected to a node their functionalities
should be taken over by the node agent (compare Fig. 3). Those
systems are further on called ‘offline systems’, because they are no
longer modelled as a distinct agent. As of now, the agents’
negotiations take place as an iterative process inside the node
agent. By this the computational effort is further reduced, which
helps in simulating a bigger amount of agents and thus enhances
SIMONA’s capabilities, which is now able to perform
time-series-based network state analysis for bigger grids across
several voltage levels.
3.2 First results on system integration
With the improved version of SIMONA, the analysis simulations to
investigate the grid performance are executed. To do so, the input
parameters of stage one are bundled to several future scenarios
that should be investigated. These scenarios and their parameters
are mainly based on different forecasts to ensure a range of likely
Fig. 3 Concept of node agent with offline systems
Fig. 4 Blueprints of geographical HV-grid views
Fig. 5 Schematic overview of the four-stage approach
CIRED, Open Access Proc. J., 2017, Vol. 2017, Iss. 1, pp. 2312–2315
2314 This is an open access article published by the IET under the Creative Commons
Attribution License (http://creativecommons.org/licenses/by/3.0/)
future developments (e.g. [10,11]). As a result of the simulation,
different time series of the grid utilisation for each scenario for
several years are generated. In the next step, these time series are
taken to detect grid elements that will face critical situations in the
future [e.g. violation of the (n−1)-criterion or an overload of a
transformer]. Not only the usage of time series identifies the
critical situations, it also allows to determine the number of hours
a violation of operational requirements and their exact occurrence.
The result of this so-called hotspot analysis is then used to
optimise the existing grid.
As a third step of the proposed approach, a fitness restoration of
the investigated grid is performed. To do so, the determined
hotspots (critical grid elements) are investigated in detail and a
problem analysis of the critical situation is performed. Then, a
predefined catalogue of measures is used to remove all critical
situations. These predefined actions are a user-specific catalogue of
operational (e.g. curtailment of the feed of renewable energy
supply(-ies) (RES)) as well as structural [e.g. addition of
(innovative) assets] possibilities to improve fitness of the grid.
They have to be defined at the beginning of the optimisation process.
The described optimisation process not only restores the fitness of
the grid but also estimates the expected costs resulting from the
different operational and structural actions. Hence, an optimisation
problem with different restrictions (e.g. maximal costs allowed) is
formulated and has to be solved.
As an outcome of the steps two and three, the network planner
gets a full overview over the different scenarios, their impact on
the grid utilisation, possible issues that might occur in the
investigated grid in the next years as well as a possible solution to
prevent them. Furthermore, information about the estimated costs,
depending on the considered scenario and the different operational
actions and structural changes. These results will be accessible via
a user interface which allows analytical visualisation of simulated
data even in big grids with large volumes of measures (e.g. load
factors) in a multidimensional fashion. Technologically, this is
achieved by use of GeoJSON, Openstreetmap, Java and a Model
View ViewModel concept in a four-tier architecture. To ensure
analytical efficiency, the database layer provides NewSQL features
such as in-memory computing and columnar organisation. A first
mock-up of a geographical representation is shown in Fig. 4.
The values of resulting measures are represented directly using
appropriate colour schemes or accessible in dedicated views (e.g.
load duration curves for individual assets). A schematic overview
on the described four-stage approach is presented in Fig. 5.
4 Conclusion
This paper presents an approach to enhance the distribution grid
planning process while taking into account multiple-voltage levels.
With the widely discussed emerging volatility of grid loading and
increasing (market oriented) interaction of distribution grid
participants, the need for sophisticated, cross-voltage grid-state
analysis arises.
Such a detailed analysis can be carried out by a time-series-based
MAS, which will produce a huge amount of data from which
information have to be extracted. To achieve an optimal
distribution grid planning process, those resulting information have
to be processed further, to derive adequate measures. Those steps
can hardly be performed in a manual way.
The project ‘Agent.GridPlan’develops an approach for assisting
distribution grid planners by integrating the agent-based grid
analysis and simulation tool SIMONA and an optimised
decision-making process into an interactive system. The tool is
either capable of deriving adequate measures from sophisticated
grid-state analysis, as well as of giving a powerful visualisation
module. Hence, planning engineers would be able to understand
given complex interdependencies and their resulting consequences.
The developed approach enables network planners to improve
their network planning process due to:
†multiple-voltage-level analysis in one step,
†consideration of different scenarios over multiple years,
†automatic optimisation of the critical situations in the grid via a
predefined measure catalogue with operational and structural actions,
†estimated costs of the different actions depending on the scenario
and its occurrence which allows a financial evaluation as well as and
†a state-of-the-art and user-friendly interface.
5 Outlook
With the systems’integration and functionality enhancement some
major challenges arise. With emerging voltage levels, the number
of network participants has to be taken into account. Thus, the
bidirectional interaction between market-oriented load and feed-in
systems with a wholesale market needs to be represented. The
higher amount of individuals that have to be simulated also
challenges computational and data efficiency. Here, adequate
solutions have to be found in order to keep the simulation time
and computational requirements in an acceptable size. As already
mentioned, the planning engineer is challenged in analysing the
gathered information. Here, a suitable solution for human machine
interaction has to be found.
6 Acknowledgments
This project was supported by the European Regional Development
Fund (ERDF). See also http://www.agent-gridplan.net.
7 References
1 Jendernalik, L., Giavarra, D., Engels, C., et al.: ‘Resilient and comprehensive
network planning is key to a successful integration of renewable energy
resources’. CIRED, Helsinki, 2016
2 Engels, C., Jendernalik, L., von Haebler, J., et al.: ‘Smart planning –optimal
balance between risk & costs’. 23rd CIRED Int. Conf., Lyon, 2015
3 Kays, J., Seack, A., Häger, U.: ‘The potential of using generated time series in the
distribution grid planning process’. CIRED, Lyon, 2015
4 McArthur, S.D.J., Davidson, E.W., Catterson, V.M., et al.: ‘Multi-agent systems
for power engineering applications—part I: concepts, approaches, and technical
challenges’,IEEE Trans. Power Syst., 2007, 22, (4), pp. 1743–1752
5‘JADE v4.4.0.’Available at http://jade.tilab.com/, accessed: 12 December 2016
6 Cortese, E., Quarta, F., Vitaglione, G.: ‘Scalability and performance of JADE
message transport system’. AAMAS Work AgentCities, Bol., 2002, vol. 16,
pp. 52–65
7 Kays, J., Rehtanz, C.: ‘Planning process for distribution grids based on flexibly
generated time series considering RES, DSM and storages’,IET. Gener. Transm.
Distrib., 2016, 10, (14), pp. 3405–3412
8 Kays, J., Seack, A., Rehtanz, C.: ‘Consideration of smart-meter measurements in a
multi-agent simulation environment for improving distribution grid planning’.
Proc. IEEE Power & Energy Society Innovative Smart Grid Technologies Conf.
(ISGT), 2016, pp. 1–5
9 Seack, A., Kays, J., Rehtanz, C.: ‘Time series based distribution grid planning
approach with decentralised voltage regulation’. Proc. Power Systems
Computation Conf., Wroclaw, 2014
10 German Federal Ministry for Economic Affairs and Energy: ‘Entwicklung der
energiemärkte –energiereferenzprognose’, Berlin, June 2014
11 ‘50 Hz transmission, Amprion, TenneT TSO, TransnetBW: ‘Szenariorahmen für
die Netzentwicklungspläne Strom 2030 –Entwurf der
Übertragungsnetzbetreiber’, January 2016
CIRED, Open Access Proc. J., 2017, Vol. 2017, Iss. 1, pp. 2312–2315
2315This is an open access article published by the IET under the Creative Commons
Attribution License (http://creativecommons.org/licenses/by/3.0/)
