Verification of Ancillary Services in Large Scale Power System
Verification of Ancillary Services in Large Scale Power System
Verification of Ancillary Services in Large Scale Power System
1
2 Real-Time Hardware-In-the-Loop Framework
2.1 Overview
Verification of ancillary services in large scale power system is conducted using the Laboratory facilities
available at AAU-ET partner, namely the Smart Energy Systems (SES) Laboratory [1]. The structure of this
platform is given in Figure 1.
2
minimization, etc. This industrial controller is getting information from the downstream assets placed in
medium and low voltage networks as highlighted in the next sections of this paper. A dedicated industrial
controller (Low Voltage Grid Controller) is used to host new control functionalities in secondary substations.
This platform offers the possibility to implement and verify new control and operational strategies for flexible
assets in low voltage grids. Another industrial controller is hosting typical control functionalities implemented
in renewable based generation plants such as wind or PV. Implementation of controls in all platforms is done
using MATLAB/Simulink. New actors and/or functionalities can be easily added to the platform.
Real-Time Digital Simulator.
The Real-Time Digital Simulator is implementing a large scale energy network including systems using the
multi-domain physics approach. This means that not only the electrical network and system is captured but
also other systems such as thermal, mechanical, etc. The main goal for this Real-Time simulator is to capture
the electrical system from the transmission level (TSO) down to low voltage distribution grids (DSO). The
system is able to simulate up to 200 three-phase bus-bars in EMT domain and very large electrical networks
in RMS domain (up to 10.000 buses). Other energy systems such as district heating, gas networks and
transportation systems can also be represented. All the modelling work in this real-time platform is done in
Matlab/Simulink including Toolboxes.
Physical Assets Layer (Power Hardware-In-the-Loop)
Grid Simulator. The voltages measured in a given point in the distribution network
implemented in the Real-Time Digital Simulator are applied to a 50 KVA four-quadrant AC Supply/Sink that is
supplying a local grid where different components are connected such as DER Emulator, Flexible load, etc.
The measured current at the output of the Grid Simulator is fed in back to the Real-Time Digital Simulator.
Thus, all physical equipment connected to this local grid is “seen” as part of the large real-time model
implemented in the digital simulator.
Dispersed Energy Resource. A four quadrant power converter based emulator for dispersed
generation (±20kW/10kVAR) is used to mimic characteristics of small wind turbines, PV systems or energy
storage.
Flexible loads. Three-phase independently controllable loads are used to mimic the behavior
of household appliances. Central Energy Management System for the entire house including communication
is also implemented in a dedicated server. A smart meter is providing power and energy consumption to the
upper hierarchical control levels.
Testing of new power converter based systems is also possible using this setup.
The key features of this Laboratory are as follows:
The ICT layer in SES laboratory mainly comprises of a network emulator, traffic generator, visualization server
and a high-speed switch. This switch is a fast 1 GB VLAN enabled Ethernet switch that enables full control
over the network traffic, such that traffic between controllers, assets and other entities to the network can
be redirected through a network emulator, as illustrated in Figure 3. The switch is setup using VLAN to
forward any received packets to the network emulator, which is then setup to bridge the different VLANs.
The network emulator workstation provides other services necessary for operating the test bed, e.g. DHCP
services for test bed components, secure Internet access in the test bed, trace and log options for capturing
the network traffic via a PCAP module [3].
4
Figure 3. Real-time ICT layer with network emulator connecting grid assets and controllers.
KauNet [4] is used as the network emulator in ICT layer that enables ingoing as well as outgoing data packets
to pass a queue configured with a buffer length and service time according to a given stochastic model of a
network. KauNet provides control on bit errors, packet losses, bandwidth, and delay changes. Using KauNet,
reproducible behavior of network along with an exact control on network traffic over Internet can be
provided [2]. The traffic using KauNet is routed through a set of buffers, where each buffer emulates specific
network characteristics in terms of delay, packet losses etc. The network capacity can also be emulated by
shaping the buffer sizes, such that packet losses can be efficiently emulated. The queues can internally be
linked together if needed, e.g. if multiple types of networks are connected.
5
measurements of real events in the power grid. Thus, typically open loop testing is performed. Some of these
events i.e. large frequency excursion may not occur during the limited testing period of controllers.
In order to overcome some of the drawbacks encountered during the design and testing of control schemes
for power systems a model-based design approach can be utilized. An overview of different stages from
development to testing is given in Figure 4.
S and/or Z Dom
Desktop Simulations. Initially the control development and tuning process will use simplified/reduced order
models of power system and assets. Either continuous or discrete time models can be used according to the
tuning methodology. However, since the dynamics in the power system applications involve a wide range of
time constants and various sampling time for the subsystems involved a continuous time domain tuning is
preferred. Then, considerations about translation of models in discrete models shall be made in respect to
the actual expected sampling time of various control platform used.
Rapid Control Prototyping. In this stage the entire model is implemented in discrete domain with the proper
sampling time for various subsystems. This model is implemented in a Real-Time discrete simulator e.g. Opal-
RT. The control algorithms are tested intensively under various operating conditions to identify the
robustness of the design. Worth to mention that information exchange between different subsystems i.e.
plant controller, aggregator, assets, etc. shall be captured properly with a dedicated model that is able to
simulated the communication networks and their traffic. Based on the executed tests and their results
iterations of the control design and tuning may be necessary. At this stage it can be also identified if specific
information and communication technologies can meet the requirements imposed by control response.
Specifications about minimum requirements for ICT to provide a specific control function can also be derived
here. This will allow the selection of the ICT for the deployment. The control is ready for implementation on
the target hardware when the control design verification is finalized.
Automatic Coding Implementation. Typically, target hardware requires coding of control algorithms in
different software specific to manufacturers of the platforms e.g. structured text, C/C++, etc. This process
can be shortened if the simulation tools used for control design are capable of generating automatically the
6
code such as Matlab/Simulink. Thus, this stage can be skipped when using the facilities available in Smart
Energy Systems Laboratory.
RT-HIL Co-Simulation. Verifying control algorithms on large scale power systems requires testing of the
controller platform connected to a Real-Time model of the power system including assets i.e. wind power
plants, PV plants, energy storage devices, etc. Moreover, for a realistic testing a real-Time model of the
communication networks shall be used. Thus, the controller platform including the developed algorithms in
the first phase can be tested in realistic condition close to daily operation. Power system events that cannot
be measured in real life can also be replicated in a controlled environment while data traffic associated to
specific network technologies is captured properly without actually involving the real technologies. The
advantage of this approach can be summarized as:
• Extensive testing and verification of control algorithms under operating conditions which cannot be
encountered during the field-test trials.
• Accounting for impact of communication technologies, protocols and traffic on the developed
control algorithms in a unified and consistent without being constraint by the physical systems.
The developed control algorithms including physical implementation on target hardware is then ready for
site testing. This stage corresponds to a Technology Readiness Level of 6 according to [2].
Validation. The actual controller platform is tested on-site under operating conditions allowed by the
physical power grid and assets. Typically, the targeted tests have to be reported and documented to power
system operator (TSO or DSO) as well as to the plant operator or owner before execution. This process may
be complicated when different owners/operators are involved. The main goal for this process is to ensure
that no physical equipment will be damaged during the tests. The testing campaign is typically limited in time
and power system events in scope for the developed algorithms e.g. large voltage and frequency excursions
may not occur in the system during this period. Thus, an open loop approach is used. This means that the
controller is fed with pseudo-measurements and the output of the plants is recorded. However, the actual
impact on the power grid cannot be evaluated as well as possible control interactions between assets. These
recordings may be used to validate some of the developed models including modelling assumptions used in
the previous stages.
The existing facilities in Smart Energy Systems Laboratory allows all the above design and verification
procedure being a powerful environment for achieving high TRLs.
2.4 Summary
The Real-Time Hardware-In-the-Loop framework used for validation of ancillary services in RePlan is briefly
presented. The mechanisms for emulating in Real-Time the communication networks and traffic is detailed.
A new methodology for designing and verifying controls in power systems is proposed.
7
3 Validation of ICT Model
3.1 Background
The integration of ICT into the electrical energy infrastructure is shifting from a phase of demonstration to
large-scale deployment. This will have a strong impact on system architectures as well as it raises concerns
about cyber-security issue. The electric power grid integrated with communication systems and renewable
energy source (such as solar, wind, etc.) has become a cyber-physical energy system, also termed as smart
grids. A general framework for smart grids is therefore required for the validation which takes into account
their mutual interactions and interdependencies. One of the main barriers to this has been the lack of design
and validation tools that are capable of analyzing power and communication systems in a holistic manner.
Extending power system simulation tools for the ICT domain, or vice versa, demands a lot of effort and
collaboration among experts of both areas, because the life cycle and technical specifications of the electrical
and communication equipment (in terms of reliability requirements, round trip time, determinism, temporal
consistency and hierarchy) are significantly different [5]. By creating a so-called co-simulation environment
for the integrated analysis of both domains, via ad-hoc connections (or in a master/slave fashion), one can
understand the impacts of different communication solutions used for the operation of power systems much
better. Although simulation architectures may vary, a co-simulation framework allows in general the joint
and simultaneous investigation of models developed with different tools, in which the intermediate results
are exchanged during the execution of the simulations. However, the sub-systems are usually solved
independently by their corresponding domain-specific simulators [5]. Co-simulation allows to have a
complete view of both network behavior and the physical energy system states, while power system and
communication networks are simulated with the most suitable solver and the calculation loads are shared.
In the following, two test simulation setups (off-line and on-line) are implemented where a same input signal
is sent and received with different network conditions. Main purpose of the two test setups is to validate the
off-line communication model with its on-line counterpart.
8
3.3 Setup 2- RT Online Model
In this setup, a signal generated at one working station was sent to the second working station through
KauNet network emulator under the same network conditions as in off-line test setup. The original signal as
well as the delayed signals were captured and recorded on working station 2, as shown in Figure 6.
9
effected, for instance, packet drops etc. Specifically, while considering public networks as a means of
coordination between ReGen plants and the system operators (TSO/DSO), a constant delay or packet loss
cannot be guaranteed. It is, therefore, necessary to see how different network conditions (in terms of higher
delays or packet drops) affect the performance of a signal. Thus, for both off-line as well on-line test setup,
the original signal was sent through the network under two conditions i.e.:
1. Added delays with zero packet loss rate in the network
2. Added packet loss rate with constant delay in the network
It is pertinent to mention here that the delays added in the network in case 1 are not constant delays. These
delays are rather based on delay patterns with mean values of 10ms, 100ms, 250ms and 500ms. The delays
with such mean values were considered based on the delay patterns obtained from a real network in [6] and
[7]. The real network based delay patterns in [6] were used as a base case in Deliverable D2 [7].
Here, the delay and packet drops are divided into four groups to get a better understanding of four different
network conditions, as given in Table 1:
10
Figure 8. Original Square Wave used in Off-line and On-line tests to validate impact of delay on a signal.
Based on the average delays and packet drop rates used to send these original signal through a network,
there are four test cases defined for both test setups (off-line + on-line), as shown in Table 2
11
Figure 9. Test Results from off-line setup showing received sine wave signals obtained through a network with four different
average delay patterns.
Figure 10 shows the received square wave signals from network emulator under the same delay based test
cases. It can be observed how a signal can be affected due to different (average) delays in a network.
12
Figure 10. Test Results from off-line setup showing received square wave signals obtained through a network with four different
average delay patterns.
13
Figure 11. Test Results from off-line setup showing received sine wave signals obtained through a network with four different
packet loss rates.
In order to have a better understanding and proper visualization of increased packet loss rates in a network,
the original square wave is changed to the one shown below (see Figure 12). The period of this square wave
is changed to 0.1 sec.
14
Figure 12. Original Square Wave used in Off-line + On-line tests to validate impact of packet loss on a signal.
Figure 13. Test Results from off-line setup showing received square wave signals obtained through a network with four different
packet loss rates.
15
3.6.2 Online Setup – Test Results
Figure 14. Test Results from RT online setup showing received signals obtained through a network with four different average
delay patterns.
Figure 15 captures the original sine wave with the ones received from offline as well as online test setups
(with avg. delay = 500 msec.) to show that the results obtained from both test setups are comparable.
Figure 15. Original sine wave signal compared with delayed signals received from off-line and online test setups.
16
Figure 16 shows the received (delayed) square wave signal from working station 1 through KauNet network
emulator under the four delay based test cases. As in offline results, it can be clearly observed how the sent
signal is affected due to different (average) delays in a network.
Figure 16. Test Results from RT online setup showing received signals obtained through a network with four different average
delay patterns.
17
Figure 17. Test Results from RT online setup showing received sine wave signals obtained through a network with four different
packet loss rates.
Figure 18. Test Results from RT online setup showing received square signals obtained through a network with four different
packet loss rates.
18
3.7 Results Discussion and Summary
From the above two test setups (off-line and on-line), same input signal was sent and received with different
network conditions. The results obtained from online test setup clearly validate the results from the offline
test setup. However, as a conclusion it is important to mention here that the electric power devices do not
have communication capability by themselves. Each electric device is attached with an embedded computer
system to serve as the communication interface to the network infrastructure. The electric device and the
embedded computer system together form an IED. The message processing steps within an IED are illustrated
in Figure 19, in which a message containing the device status data is generated and transmitted through four
modules in the IED i.e.:
(i) analog–digital converter transforms a status measurement into digital data,
(ii) CPU processes the measurement data,
(iii) set-point structure stores the current measurement data, and
(iv) network protocol stack formats the message and sends it into the network.
As soon as the message is received at the other (receiving) end, it passes through the same four modules.
The time spent within IEDs is part of the end-to-end delay that effects the whole control cycle.
Based on the time spent in the whole process, a control cycle can be divided into four periods, i.e.:
(i) Read – time taken to read the incoming signal,
(ii) Compute – after reading the signal, the controller computes the set-points to be sent,
(iii) Send – the controller sends the set-points, and
(iv) Finally, the controller gets ready for the next incoming signal.
A signal(s) delayed or lost during this cycle will result in wrong/invalid computation of the set-point values,
especially in case where the delay increases sampling time, as shown in Figure 21.
19
Figure 21. Message signal diagram showing a scenario where delay increases the sampling time.
These scenarios can also be observed in the above test cases, for instance, in case the average delay increases
up to 500 ms and/or packet loss rate is increased up to 90% due to poor network conditions. The two
scenarios are shown in Figure 22 and Figure 23, respectively. Therefore, it is necessary to send a right set-
point value at the right time, while making sure that it is not lost/dropped on the way.
Figure 22. Original square wave signal compared with the signal delayed above sampling time.
20
Figure 23. Original square wave signal compared with the signal received under poor network conditions (higher packet loss
rates).
Overall, the response of the off-line ICT model is matching the one provided by the RT Network Emulator
under the same input and parameters. This off-line model proved to be a powerful but yet simple
representations of the communication networks and their traffic especially for off-line multi-run studies
focusing on verification of control design. However, in order to have complete confidence selected test cases
shall go to a validation stage in the RT HIL framework. Notice that in this case longer simulation time is
expected as the entire system is running in Real-Time.
21
4 Validation of Distributed On-Line Voltage Coordination
4.1 Control Architecture
The control architecture for Distributed On-Line Voltage Coordination as defined in [7] is given in Figure 24.
HV
Vstp , droop
P + jQ
Vmeas
It requires grid layout and parameters provided by the DSOs, measurements from secondary side of primary
substations as well as measurements from each of the controlled ReGen plants. It provides voltage setpoints
and droop values for each ReGen plant considered in the asset’s portfolio.
22
Host PC-Opal RT is a computer dedicated to development of the RT model but also to monitor and to control
the RT Model using a dedicated TCP/IP link. Automatic code generation of the RT Model is included in the
RT-Lab suite provided with the Opal-RT system [10]. The Monitoring and Control Module is sending the
following signals to the Opal-RT System:
• Meteorological data: wind speeds, solar irradiation
• External grid voltage setpoint
• Load profiles in terms of active and reactive values
This module is also receiving selected signal from the Opal-RT System for monitoring the model performance
during simulations.
All the data exchange between Aggregator Hardware and Opal-RT system is executed through the RT ICT
Emulator. This means that only these signals are prone to delays, packet drop and cyber security threats. The
data exchange between Host PCs and corresponding hardware is not affected.
Vstp , droop HV
MV
ReGen P + jQ
Plant 1 Vmeas
Vstp , droop PoC1
P + jQ
Vmeas
Figure 25. RT-HIL setup used for validation of Distributed On-Line Voltage Coordination.
23
4.2.2 Aggregator Control
The Distributed On-Line Voltage Coordination presented and used in WP2 is used as such. The only
modifications are related to TCP/IP interfaces between different hardware platforms.
24
3. PVP 2 & 3 (2.5 MW each) representing typical rooftop systems mounted on top of large industrial
plants and shopping centers
Based on the control architecture described in Section 4.1, this structure of the BDG is used to analyze voltage
control in time domain for a volatile power profile of the ReGen plants, used as a benchmark test scenario
that covers the crucial operating points with high solar irradiation and high wind speed (see [7]). The ReGen
plants models are implemented in MATLAB/Simulink and the phasor model of the BDG is implemented in
OpalRT’s real-time simulation platform ePHASORsim. In [7], this model was used for off-line simulations,
while in this chapter on-line validation including the ICT layer will be performed to validate off-line simulation
results.
Two options for on-line coordination were taken into consideration in [7]:
• Updating the voltage droop values
• Updating the voltage set-point of the ReGen plants
However, it was ascertained in [7] that updating the droop settings according to the actual operating point
of ReGen plants did not significantly reduce the power losses within the distribution grid. While, updating
the voltage set-points according to the actual operating point of ReGen plants reduced the power losses
within the distribution grid to a measurable extent. Moreover, it decreased the reactive power utilization
rate of the ReGen plants, in comparison to Control Concept 2 (Distributed Off-Line Coordination) in [7].
Therefore, in this section we only consider the case of updating the voltage set-points of the ReGen plants.
In [7], four different update rates (10 sec., 1 min., 5 min., 15 min.) were considered in simulations for
adjusting the voltage set-point of each ReGen plant to evaluate their impact on the power losses within the
grid. Since it has already been shown in Chapter 3 that the offline communication model performs the same
as the online communication setup (with KauNet), two update rates are considered as test cases to validate
the offline simulations via online setup:
25
1. 𝑇𝑇𝑠𝑠 = 10 sec.
2. 𝑇𝑇𝑠𝑠 = 1 min.
A time frame of one hour is considered sufficient to represent a volatile power profile covering the extreme
operational points at high wind speed and solar irradiation.
Figure 27. P, Q, V of PVP 1 over one hour for 10 sec. update rate.
26
Figure 28. P, Q, V of PVP 2 over one hour for 10 sec. update rate.
Figure 29. P, Q, V of PVP 3 over one hour for 10 sec. update rate.
27
Figure 30. P, Q, V of WPP over one hour for 10 sec. update rate.
Figure 31. P, Q, V of PVP 1 over one hour for 1 min. update rate.
28
Figure 32. P, Q, V of PVP 2 over one hour for 1 min. update rate.
Figure 33. P, Q, V of PVP 3 over one hour for 1 min. update rate.
29
Figure 34. P, Q, V of WPP over one hour for 1 min. update rate.
It can be observed from Figure 27 to Figure 34 that no significant differences are observed for the specified
test cases in the online test setup as compared to the offline tests. However, since there is a little difference
seen in the some results, it is worth calculating the percentage error in each case. The percentage error in
case of voltage (V) and active power (P) is calculated based on the difference between offline and online
results, keeping offline results as reference:
𝑉𝑉𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜 − 𝑉𝑉𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜
𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 [%] = × 100
𝑉𝑉𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜
The same applies for the active power (P). While for the reactive power (Q), it can be observed that the signal
includes zero, which will give infinitely large errors once divided by the reference signal (offline). Therefore,
for the reactive power (Q), a delta signal (∆Q) is plotted which is the difference between the Q obtained
offline and online:
∆𝑄𝑄 = 𝑄𝑄𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜 − 𝑄𝑄𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜
30
Figure 35. Error in V, P and Q of PVP 1 sent via offline and online communication models over one hour with 10 sec. update rate.
Figure 36. Error in V, P and Q of PVP 2 sent via offline and online communication models over one hour with 10 sec. update rate.
31
Figure 37. Error in V, P and Q of PVP 3 sent via offline and online communication models over one hour with 10 sec. update rate.
Figure 38. Error in V, P and Q of WPP sent via offline and online communication models over one hour with 10 sec. update rate.
32
Figure 39. Error in V, P and Q of PVP 1 sent via offline and online communication models over one hour with 1 min. update rate.
Figure 40. Error in V, P and Q of PVP 2 sent via offline and online communication models over one hour with 1 min. update rate.
33
Figure 41. Error in V, P and Q of PVP 3 sent via offline and online communication models over one hour with 1 min. update rate.
Figure 42. Error in V, P and Q of WPP sent via offline and online communication models over one hour with 1 min. update rate.
It can be observed from Figure 35 to Figure 42 that in case of V and P from all ReGen plants, collectively the
error remained less than 1% approximately. The Ethernet and other real-time effects in the link can be a
34
reason to impose this error. While the relatively higher error initially in each case is expected due to the HIL
initialization setup.
4.5 Summary
Two relevant test cases from WP2 [7] are selected for comparison and verification. These test cases are co-
simulated in the RT HIL setup. Overall errors below 1% are obtained when comparing main variables from
off-line studies with RT HIL results. Some high errors are seen in the beginning of simulations due to
initialization of the RT system. However, these errors are becoming small after the initialization is done. The
validation of voltage control coordination is achieved using a RT HIL setup.
35
6 References
[1] »Smart Energy Systems Laboratory,« 2018. [Online]. Available:
http://www.et.aau.dk/laboratories/power-systems-laboratories/smart-energy-systems/.
[2] A. Azim og Z. I. Awan, »Network Emulation, Pattern based shaping and KauNET Evaluation,« 2008.
[3] »HORIZON 2020 – WORK PROGRAMME 2014-2015,« European Commission Decision C , 2014.
[4] J. Garcia, E. Conchon, T. Pérennou og A. Brunstrom, »KauNet: Improving Reproducibility for Wireless
and Mobile Research,« 2007.
[7] L. Petersen, F. Iov, K. Shahid, R. L. Olsen, M. Altin og A. D. Hansen, »D2 - Voltage control support and
coordination between ReGen plants in distribution systems,« 2016.
36